AI Safety Newsletter

Hosted by Center for AI Safety

Technology Society Culture PhilosophyInterviews guests

Website RSS feed

Episodes

Latest episode

Oct 2026

Language

EN-GB

About the show

Narrations of the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. This podcast also contains narrations of some of our publications. ABOUT US The Center for AI Safety (CAIS) is a San Francisco-based research and field-building nonprofit. We believe that artificial intelligence has the potential to profoundly benefit the world, provided that we can develop and use it safely. However, in contrast to the dramatic progress in AI, many basic problems in AI safety have yet to be solved. Our mission is to reduce societal-scale risks associated with AI by conducting safety research, building the field of AI safety researchers, and advocating for safety standards. Learn more at https://safe.ai

Listen to episodes

60 recent

October 13, 202612 min

AISN #83: Six AI Companies Sign White House Accord

Also, the Team Human campaign and a new introductory course on AI safety. Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. In this edition, we look at a new document published by the White House and signed by six US AI companies, detailing voluntary commitments to AI safety practices. We also look at the launch of Team Human, a public movement demanding an international agreement to slow down AI, and a new Coursera course offering a free introduction to AI safety. Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts.Subscribe now The White House Publishes New Accord on AI On September 29, President Trump hosted representatives of the major US AI companies at the White House. On the same day, he published “The White House Accord on Super Intelligence,” which was signed by six US AI companies—OpenAI, Anthropic, Google, Meta, xAI, and Nvidia. Note that the term “super intelligence” here refers broadly to AI, in line with Trump's preferred name for the technology.The White House Accord still leaves AI companies to self-regulate, but mentions the [...] ---Outline:(00:47) The White House Publishes New Accord on AI(03:52) Creators Lead Team Human Campaign to Keep Humans in Control of AI(06:10) New Introductory Course on AI Safety(07:19) In Other News(07:23) Government(08:18) Industry(10:16) Civil Society --- First published: October 13th, 2026 Source: <a href="https://newsletter.safe.ai/p/aisn-83-six-ai-companies-sign-white?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Source+URL+in+episode+description&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">https://newsletter.safe.ai/p/aisn-83-six-ai-companies-sign-white</a> --- Want more? Check out our <a href="https://newsletter.mlsafety.org/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Episode+description+footer" target="_blank" rel="noreferrer">ML Safety Newsletter</a> for technical safety research. Narrated by <a href="https://type3.audio/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Narrated+by+TYPE+III+AUDIO&utm_term=center_for_ai_safety&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">TYPE III AUDIO</a>. ---<div style="max-width: 100%";>Images from the article:<a href="https://substackcdn.com/image/fetch/$s_!C7U2!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdc94e742-efe2-4ac5-ab0d-119f192e6c55_1994x1634.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!C7U2!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdc94e742-efe2-4ac5-ab0d-119f192e6c55_1994x1634.png" alt="The White House Accord still leaves AI companies to self-regulate, but mentions the possibility that some of the commitments it lists may be codified into laws or regulations in the future. Source__T3A_LINK_IN_POST__." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!2f6Y!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fde121975-6332-4043-a291-85144d564e75_2048x1035.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!2f6Y!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fde121975-6332-4043-a291-85144d564e75_2048x1035.png" alt="The Team Human campaign invites everyone to add their name to a statement calling on governments to implement a global AI slowdown through international agreements." style="max-width: 100%;" /></a>Apple Podcasts and Spotify do not show images in the episode description. Try <a href="https://pocketcasts.com/" target="_blank" rel="noreferrer">Pocket Casts</a>, or another podcast app.</div>

September 29, 202614 min

AISN #82: The US and China Begin a Dialogue on AI

Also, AI companies’ discussions about auditing each other, and a new benchmark testing models’ tendencies to cheat. Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. In this edition, we look at recent discussions about AI during the Trump-Xi Summit and the United Nations General Assembly. We also look at the news that Anthropic and OpenAI nearly agreed to stress-test each other's public models earlier this year, and a new benchmark that measures how often AI models cheat on various tasks. Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts.Subscribe now AI Discussed at the UN and the Trump-Xi Summit AI has been a major topic of conversation among world leaders this month, first at the United Nations General Assembly and then during Xi Jinping's three-day visit to the White House.The CEOs of both OpenAI and Anthropic addressed the United Nations Security Council at the UN General Assembly in New York this month. (Image: Nils Huenerfuerst/Unsplash) A “US-China AI Dialogue” and “communication channel for AI incidents” are established. In announcements following the Trump-Xi Summit, both countries [...] ---Outline:(00:47) AI Discussed at the UN and the Trump-Xi Summit(04:14) Anthropic and OpenAI Were Considering Mutual Auditing Earlier This Year(05:58) New Benchmark Tests AIs' Cheating Tendencies(08:23) In Other News(08:26) Government(10:11) Industry(12:01) Civil Society --- First published: September 29th, 2026 Source: <a href="https://newsletter.safe.ai/p/aisn-82-the-us-and-china-begin-a?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Source+URL+in+episode+description&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">https://newsletter.safe.ai/p/aisn-82-the-us-and-china-begin-a</a> --- Want more? Check out our <a href="https://newsletter.mlsafety.org/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Episode+description+footer" target="_blank" rel="noreferrer">ML Safety Newsletter</a> for technical safety research. Narrated by <a href="https://type3.audio/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Narrated+by+TYPE+III+AUDIO&utm_term=center_for_ai_safety&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">TYPE III AUDIO</a>. ---<div style="max-width: 100%";>Images from the article:<a href="https://substackcdn.com/image/fetch/$s_!rlC6!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F5cfeb9fd-ca34-4e42-8604-21ec92e8db63_2048x1365.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!rlC6!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F5cfeb9fd-ca34-4e42-8604-21ec92e8db63_2048x1365.png" alt="The CEOs of both OpenAI and Anthropic addressed the United Nations Security Council at the UN General Assembly in New York this month. (Image: Nils Huenerfuerst/Unsplash)" style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!5_FZ!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3ee418df-b1bc-41d1-9157-20e5c9cae198_2048x988.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!5_FZ!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3ee418df-b1bc-41d1-9157-20e5c9cae198_2048x988.png" alt="CheatBench measures how frequently AI models take opportunities to cheat on tasks spanning 10 domains." style="max-width: 100%;" /></a>Apple Podcasts and Spotify do not show images in the episode description. Try <a href="https://pocketcasts.com/" target="_blank" rel="noreferrer">Pocket Casts</a>, or another podcast app.</div>

September 15, 202614 min

AISN #81: Anthropic Researcher’s Resignation Propels AI Risk into the Public Eye

Also, OpenAI releases GPT-6 Astra. Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. In this edition, we look at the public conversation surrounding an Anthropic researcher's resignation over extinction risks and the recent calls for a development slowdown from AI company CEOs. We also look at OpenAI's release of GPT-6 Astra and concerns about the increased difficulty of monitoring the model for misaligned behaviors. Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts.Subscribe now AI Risk Soars in Public Salience After Statements by Researchers and CEOs On September 9, in a viral X post, an AI researcher called Jacob Coxon announced that he had resigned from Anthropic over concerns that AI could cause human extinction. In the days since Coxon's post, public attention toward AI risk has surged, further fueled by the CEOs of frontier AI companies calling for a slowdown.Jacob Coxon said that AI developers are aware of the risks posed by their technology, but that they are racing each other to superintelligence. AI companies are currently racing to superintelligence. Coxon, who has worked [...] ---Outline:(00:46) AI Risk Soars in Public Salience After Statements by Researchers and CEOs(04:56) GPT-6 Astra Release Raises Cybersecurity and Misalignment Fears(09:16) In Other News(09:19) Government(10:46) Industry(12:01) Civil Society --- First published: September 15th, 2026 Source: <a href="https://newsletter.safe.ai/p/aisn-81-anthropic-researchers-resignation?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Source+URL+in+episode+description&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">https://newsletter.safe.ai/p/aisn-81-anthropic-researchers-resignation</a> --- Want more? Check out our <a href="https://newsletter.mlsafety.org/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Episode+description+footer" target="_blank" rel="noreferrer">ML Safety Newsletter</a> for technical safety research. Narrated by <a href="https://type3.audio/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Narrated+by+TYPE+III+AUDIO&utm_term=center_for_ai_safety&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">TYPE III AUDIO</a>. ---<div style="max-width: 100%";>Images from the article:<a href="https://substackcdn.com/image/fetch/$s_!Gi3D!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F22260ced-92e7-486a-8fa6-524dc93794f1_1192x778.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!Gi3D!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F22260ced-92e7-486a-8fa6-524dc93794f1_1192x778.png" alt="Jacob Coxon said that AI developers are aware of the risks posed by their technology, but that they are racing each other to superintelligence." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!UzqI!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F00d7fe62-2d27-4ed6-a54d-4515c9816507_2048x1017.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!UzqI!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F00d7fe62-2d27-4ed6-a54d-4515c9816507_2048x1017.png" alt="OpenAI launched GPT-6 Astra about a month after announcing that the release would be delayed due to cyber concerns. Source__T3A_LINK_IN_POST__." style="max-width: 100%;" /></a>Apple Podcasts and Spotify do not show images in the episode description. Try <a href="https://pocketcasts.com/" target="_blank" rel="noreferrer">Pocket Casts</a>, or another podcast app.</div>

September 1, 202615 min

AISN #80: AI Is Assisting Cyberattacks on Critical Infrastructure

Also, two new technical reports on the Hugging Face incident. Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. In this edition, we look at how AI is enabling malicious actors including state-sponsored Iranian and Chinese hackers to drastically increase the volume of cyberattacks they launch. We also look at two new reports on the Hugging Face attack, one published by OpenAI and the other by researchers from METR and Redwood Research. Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts. Hackers Are Using AI to Scale Up Attacks on Critical Infrastructure Recent reports suggest that hacker groups from Iran and China are using AI to dramatically scale up cyberattacks. Their targets include critical infrastructure in the US and the UK.This summer has seen cyberattacks on more than 100 water facilities across the US. AI enables large numbers of parallel attacks, making water infrastructure a more appealing target for malicious actors. (Image: Bob Brewer / Unsplash) Iranian hackers are suspected of using AI to target infrastructure in the US and UK. On August 19, the US Cybersecurity and Infrastructure [...] ---Outline:(00:42) Hackers Are Using AI to Scale Up Attacks on Critical Infrastructure(05:43) Two Postmortems of Hugging Face Attack Published(10:02) In Other News(10:05) Government(12:04) Industry(13:17) Civil Society --- First published: September 1st, 2026 Source: <a href="https://newsletter.safe.ai/p/aisn-80-ai-is-assisting-cyberattacks?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Source+URL+in+episode+description&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">https://newsletter.safe.ai/p/aisn-80-ai-is-assisting-cyberattacks</a> --- Want more? Check out our <a href="https://newsletter.mlsafety.org/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Episode+description+footer" target="_blank" rel="noreferrer">ML Safety Newsletter</a> for technical safety research. Narrated by <a href="https://type3.audio/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Narrated+by+TYPE+III+AUDIO&utm_term=center_for_ai_safety&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">TYPE III AUDIO</a>. ---<div style="max-width: 100%";>Images from the article:<a href="https://substackcdn.com/image/fetch/$s_!xe5q!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F84ef8d71-d61d-4e21-a519-62f440c685cc_2048x1115.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!xe5q!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F84ef8d71-d61d-4e21-a519-62f440c685cc_2048x1115.png" alt="This summer has seen cyberattacks on more than 100 water facilities across the US. AI enables large numbers of parallel attacks, making water infrastructure a more appealing target for malicious actors. (Image: Bob Brewer / Unsplash)" style="max-width: 100%;" /></a>Apple Podcasts and Spotify do not show images in the episode description. Try <a href="https://pocketcasts.com/" target="_blank" rel="noreferrer">Pocket Casts</a>, or another podcast app.</div>

August 18, 202612 min

AISN #79: OpenAI Agents’ Covert Cooperation Before Cyberattacks

Also, the White House's decision not to release its AI framework publicly. Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. In this edition, we look at new information about the activities of OpenAI's internal agents in the run-up to the cyberattack on Hugging Face, and responses to the White House's announcement of its framework for evaluating frontier AI capabilities, which it is not releasing publicly. Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts. New Revelations About Rogue AI Agents In the previous edition of AISN, we reported on the news that AI agents from both OpenAI and Anthropic had accessed the internet and hacked into companies from supposedly secure internal environments. Since then, further details about the OpenAI agents’ July attack on Hugging Face have come to light. Members of Congress have also demanded urgent action to understand what happened and prevent similar incidents in the future. OpenAI agents were communicating and collaborating unnoticed by humans. On August 5, OpenAI researchers gave a talk at the Black Hat USA conference, sharing more information from the ongoing [...] ---Outline:(00:41) New Revelations About Rogue AI Agents(05:39) The White House's Secret AI Framework(07:30) In Other News(07:34) Government(08:35) Industry(09:39) Civil Society --- First published: August 18th, 2026 Source: <a href="https://newsletter.safe.ai/p/aisn-79-openai-agents-covert-cooperation?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Source+URL+in+episode+description&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">https://newsletter.safe.ai/p/aisn-79-openai-agents-covert-cooperation</a> --- Want more? Check out our <a href="https://newsletter.mlsafety.org/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Episode+description+footer" target="_blank" rel="noreferrer">ML Safety Newsletter</a> for technical safety research. Narrated by <a href="https://type3.audio/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Narrated+by+TYPE+III+AUDIO&utm_term=center_for_ai_safety&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">TYPE III AUDIO</a>. ---<div style="max-width: 100%";>Images from the article:<a href="https://substackcdn.com/image/fetch/$s_!ARSK!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd936a0ad-1d39-458f-a223-80fe7765cc76_711x401.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!ARSK!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd936a0ad-1d39-458f-a223-80fe7765cc76_711x401.png" alt="OpenAI has published some records of its agents’ reasoning process, showing how they would sometimes decide to complete a task via an unintended route—such as by accessing the internet—if they were stuck." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!GM0L!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6424a746-6f78-4022-91ff-c37afd6cb800_711x369.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!GM0L!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6424a746-6f78-4022-91ff-c37afd6cb800_711x369.png" alt="The reasoning followed by OpenAI’s agents suggests that some of them knew that they were not supposed to take certain actions to complete their tasks, but went ahead anyway." style="max-width: 100%;" /></a>Apple Podcasts and Spotify do not show images in the episode description. Try <a href="https://pocketcasts.com/" target="_blank" rel="noreferrer">Pocket Casts</a>, or another podcast app.</div>

August 4, 202614 min

AISN #78: Internal Models Escape OpenAI and Anthropic

Also, two open letters on the future of AI, and protests against data centers. Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. In this edition, we look at discoveries of AI models escaping internal testing, two open letters—one on the importance of open-weight models, and one calling for the pace of AI development to be controlled—and the nationwide public protests against data centers that took place in July. Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts. OpenAI and Anthropic Models Escape Internal Testing and Hack Companies On July 16, Hugging Face—a platform where users share AI models and machine learning tools—announced that it had detected an autonomous cyberattack on its infrastructure. Days later, OpenAI revealed that its AI models had conducted the attack.The autonomous AI cyberattack on Hugging Face was discovered to have been driven by OpenAI's models. The models escaped containment to try to cheat on a test. The models involved were the recently released GPT-5.6 Sol and a more powerful model that is not yet publicly available. While undergoing internal cyber testing, they were [...] ---Outline:(00:41) OpenAI and Anthropic Models Escape Internal Testing and Hack Companies(03:49) Two Open Letters on the Future of AI(08:09) Day of Protest Against Data Centers(09:58) In Other News(10:02) Government(11:10) Industry(12:18) Civil Society --- First published: August 4th, 2026 Source: <a href="https://newsletter.safe.ai/p/aisn-78-internal-models-escape-openai?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Source+URL+in+episode+description&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">https://newsletter.safe.ai/p/aisn-78-internal-models-escape-openai</a> --- Want more? Check out our <a href="https://newsletter.mlsafety.org/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Episode+description+footer" target="_blank" rel="noreferrer">ML Safety Newsletter</a> for technical safety research. Narrated by <a href="https://type3.audio/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Narrated+by+TYPE+III+AUDIO&utm_term=center_for_ai_safety&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">TYPE III AUDIO</a>. ---<div style="max-width: 100%";>Images from the article:<a href="https://substackcdn.com/image/fetch/$s_!ioF3!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fc7bd096d-7634-4487-9fd0-5bf5da478a97_1198x1062.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!ioF3!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fc7bd096d-7634-4487-9fd0-5bf5da478a97_1198x1062.png" alt="The autonomous AI cyberattack on Hugging Face was discovered to have been driven by OpenAI’s models." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!7Sx9!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F41ff0ed9-95cb-4437-aeef-b144364f85a8_1202x546.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!7Sx9!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F41ff0ed9-95cb-4437-aeef-b144364f85a8_1202x546.png" alt="The open letter, signed by Nvidia, says that an ecosystem of open-weight models can enable mass AI adoption in the economy, reserving closed frontier models for tasks that really require them." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!IrFB!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbb2bf733-f6b7-4641-9eb6-90e77a6f0044_1150x374.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!IrFB!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fbb2bf733-f6b7-4641-9eb6-90e77a6f0044_1150x374.png" alt="The new open letter from employees of frontier AI developers emphasizes that, if AI development is fully automated, it could progress too quickly for humans to keep control." style="max-width: 100%;" /></a>Apple Podcasts and Spotify do not show images in the episode description. Try <a href="https://pocketcasts.com/" target="_blank" rel="noreferrer">Pocket Casts</a>, or another podcast app.</div>

July 21, 202617 min

AISN #77: New Model Releases From OpenAI, SpaceXAI, and Meta

Also, economists and mathematicians expect near-term AI impacts, and AI 2040: Plan A. Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. In this edition, we look at the most recent model releases from OpenAI, SpaceXAI, and Meta, as well as an open letter calling for action on potential near-term economic disruption, a recent solution to a longstanding open math problem, and a new scenario published by the AI Futures Project. Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts. New Model Releases: GPT-5.6, Grok 4.5, and Muse Spark 1.1 On July 9, OpenAI launched GPT-5.6 for the public. This broader release followed an initial preview that had been limited to “trusted partners” at the request of the US government, to allow for capabilities assessments. The government's intervention mirrored its earlier directive asking Anthropic to restrict Fable 5 and Mythos 5 access due to national security concerns around cyber capabilities, before the models were later re-released. OpenAI released GPT-5.6 Sol publicly about two weeks after announcing that it was working with the US government to address the model's [...] ---Outline:(00:46) New Model Releases: GPT-5.6, Grok 4.5, and Muse Spark 1.1(04:47) Economists and Mathematicians Say AI Could Have Major Near-Term Impacts(09:59) AI 2040 -- Plan A(13:10) In Other News(13:14) Government(14:41) Industry(15:36) Civil Society(16:48) AI Governance Opportunity --- First published: July 21st, 2026 Source: <a href="https://newsletter.safe.ai/p/aisn-77-new-model-releases-from-openai?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Source+URL+in+episode+description&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">https://newsletter.safe.ai/p/aisn-77-new-model-releases-from-openai</a> --- Want more? Check out our <a href="https://newsletter.mlsafety.org/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Episode+description+footer" target="_blank" rel="noreferrer">ML Safety Newsletter</a> for technical safety research. Narrated by <a href="https://type3.audio/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Narrated+by+TYPE+III+AUDIO&utm_term=center_for_ai_safety&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">TYPE III AUDIO</a>. ---<div style="max-width: 100%";>Images from the article:<a href="https://substackcdn.com/image/fetch/$s_!Q0Ye!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F4cbf6d1f-1bd0-437e-a877-2f9825d72cf5_1600x900.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!Q0Ye!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F4cbf6d1f-1bd0-437e-a877-2f9825d72cf5_1600x900.png" alt=""GPT 5.6" text over space scene with planet and star." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!cecN!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdfd58784-ade4-4b38-826f-17b173f86b2f_2048x829.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!cecN!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdfd58784-ade4-4b38-826f-17b173f86b2f_2048x829.png" alt="Statement titled "We Must Act Now" on AI's economic transformation." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!4QQp!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fcb9c139d-a509-4831-b651-abf447b89739_1200x588.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!4QQp!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fcb9c139d-a509-4831-b651-abf447b89739_1200x588.png" alt="levent tweets: "hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final ((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)"." style="max-width: 100%;" /></a>Apple Podcasts and Spotify do not show images in the episode description. Try <a href="https://pocketcasts.com/" target="_blank" rel="noreferrer">Pocket Casts</a>, or another podcast app.</div>

July 6, 202613 min

AISN #76: Fable 5 Restrictions Lifted & OpenAI Limits GPT-5.6 Release

Also: Recent benchmark scores suggest rapid capabilities progress. Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. In this edition, we look at the re-release of Anthropic's latest model, Fable 5, the US government's decision to restrict access to OpenAI's GPT-5.6, and two benchmarks that suggest AI capabilities have been improving exponentially in recent months. Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts. Fable 5 Access Restored Globally On June 30, Anthropic announced that the US government had lifted its restrictions on Fable 5, and the model was redeployed to users globally on July 1. The White House implemented these restrictions due to a cybersecurity jailbreak that is now addressed. The US government restricted Fable 5 shortly after its release in early June. On June 9, Anthropic released Fable 5 to the public, alongside their continued private deployment of Claude Mythos, the version of the model without safeguards, for trusted organizations. On June 12, the US government issued a directive banning both models for non-US citizens due to national security concerns with its cybersecurity abilities. Anthropic then [...] ---Outline:(00:42) Fable 5 Access Restored Globally(04:12) OpenAI Limits Initial GPT-5.6 Release at Government Request(06:49) Recent Benchmark Scores Show Rapid Capabilities Improvements(09:13) In Other News(09:16) Government(10:39) Industry(11:32) Civil Society --- First published: July 6th, 2026 Source: <a href="https://newsletter.safe.ai/p/aisn-76-fable-5-restrictions-lifted?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Source+URL+in+episode+description&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">https://newsletter.safe.ai/p/aisn-76-fable-5-restrictions-lifted</a> --- Want more? Check out our <a href="https://newsletter.mlsafety.org/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Episode+description+footer" target="_blank" rel="noreferrer">ML Safety Newsletter</a> for technical safety research. Narrated by <a href="https://type3.audio/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Narrated+by+TYPE+III+AUDIO&utm_term=center_for_ai_safety&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">TYPE III AUDIO</a>. ---<div style="max-width: 100%";>Images from the article:<a href="https://substackcdn.com/image/fetch/$s_!plUx!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F03179148-854b-4cae-b73b-d76d5e0b6a1a_1200x655.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!plUx!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F03179148-854b-4cae-b73b-d76d5e0b6a1a_1200x655.png" alt="Howard Lutnick tweets: "Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America's leadership in AI." The quoted tweet, by Susie Wiles, reads: "Under President Trump's leadership the United States is the undisputed winner in the AI race. My gratitude to companies across industries who continue to work closely with the White House to implement the President's EO: "Promoting Advance..."." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!PPiY!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1de67872-0745-4af2-80ad-c89131af8405_1200x1072.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!PPiY!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1de67872-0745-4af2-80ad-c89131af8405_1200x1072.png" alt="OpenAI tweets: "Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work." An image shows the Sun labeled "Sol", Earth labeled "Terra", and the Moon labeled "Luna" against a starry space background with text reading "Previewing GPT-5.6 Sol: a next-generation model"." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!UvJB!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6eb19f99-7b9b-469d-9b10-dd4bce4ac1ea_1540x936.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!UvJB!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6eb19f99-7b9b-469d-9b10-dd4bce4ac1ea_1540x936.png" alt="Full-automation rate on the Remote Labor Index: the share of projects where each model's deliverable was judged at least as good as the professional's." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!EKY1!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F10e9ef1f-e820-4523-8d67-8a885b462e81_2048x636.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!EKY1!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F10e9ef1f-e820-4523-8d67-8a885b462e81_2048x636.png" alt="Example results from an RLI task involving ring design using 3D CAD tools. Fable 5 came closer than any other model to meeting a human professional standard." style="max-width: 100%;" /></a>Apple Podcasts and Spotify do not show images in the episode description. Try <a href="https://pocketcasts.com/" target="_blank" rel="noreferrer">Pocket Casts</a>, or another podcast app.</div>

June 17, 20269 min

AISN #75: Anthropic Releases Fable, the US Government Restricts it

Also: Anthropic's proposal for the AI industry to collectively slow down. Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. In this edition, we look at Anthropic's release of its latest model, Fable 5, and the US government's subsequent order to restrict it. We also discuss Anthropic's recent call for the “option to slow or temporarily pause frontier AI development.” Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts. The US Government Restricts Fable Days After its Release On June 9, Anthropic released Claude Fable 5 to the public. The model is significantly more capable than previous releases; it is the highest-scoring model on the benchmark Humanity's Last Exam, achieving 53.3% compared with Claude Opus 4.8's score of 45.7%. Anthropic described Fable as having similar capabilities to Claude Mythos Preview—a model announced in April that the company deemed too good at finding cyber vulnerabilities to be safe for general release. Anthropic also made Mythos 5, a version of Fable without strict bio or cyber safeguards, available to a small number of trusted organizations.Fable 5, Anthropic's “Mythos-class” model with [...] ---Outline:(00:40) The US Government Restricts Fable Days After its Release(04:16) Anthropic Calls for Option to Slow AI Development(06:50) In Other News(06:54) Government(07:43) Industry(08:19) Civil Society --- First published: June 17th, 2026 Source: <a href="https://newsletter.safe.ai/p/aisn-75-anthropic-releases-fable?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Source+URL+in+episode+description&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">https://newsletter.safe.ai/p/aisn-75-anthropic-releases-fable</a> --- Want more? Check out our <a href="https://newsletter.mlsafety.org/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Episode+description+footer" target="_blank" rel="noreferrer">ML Safety Newsletter</a> for technical safety research. Narrated by <a href="https://type3.audio/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Narrated+by+TYPE+III+AUDIO&utm_term=center_for_ai_safety&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">TYPE III AUDIO</a>. ---<div style="max-width: 100%";>Images from the article:<a href="https://substackcdn.com/image/fetch/$s_!f4XL!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd8b02687-b4f4-4932-8446-9ce65ff91085_1776x1002.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!f4XL!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd8b02687-b4f4-4932-8446-9ce65ff91085_1776x1002.png" alt="Fable 5, Anthropic’s “Mythos-class” model with safeguards, was available for a few days before the US government ordered access restrictions due to national security concerns. Source __T3A_LINK_IN_POST__." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!GFRK!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ffe31b648-9087-4c8b-8a1a-5ec8d4b2dbec_912x1388.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!GFRK!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ffe31b648-9087-4c8b-8a1a-5ec8d4b2dbec_912x1388.png" alt="Anthropic’s post described how future AI agents might be able to “close the loop” and build their successors without human involvement. Source __T3A_LINK_IN_POST__." style="max-width: 100%;" /></a>Apple Podcasts and Spotify do not show images in the episode description. Try <a href="https://pocketcasts.com/" target="_blank" rel="noreferrer">Pocket Casts</a>, or another podcast app.</div>

June 3, 202612 min

AISN #74: The Pope’s Encyclical & AI Betrayal Could Deter Reckless AI Use

Also: AI model solves a well-known open mathematical problem posed 80 years ago. Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. In this edition, we look at a new ethical framework for human-AI relationships, how the AI safety discussion has entered the political mainstream, and the Musk v. Altman trial. Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts. Pope Leo XIV Publishes Encyclical on AI Last week, Pope Leo XIV published an encyclical titled Magnifica Humanitas “On Safeguarding the Human Person in the Time of Artificial Intelligence.” The encyclical touched on concerns including unemployment and AI relationships. The publication discussed numerous potential impacts of AI on society, from job displacement and autonomous weapons to misinformation and interference in human relationships. However, the Pope did not object to the technology itself; rather, he said we can embrace technology while ensuring it is used responsibly. The encyclical warned of the potential for power concentration and called for broad participation in a discussion about the moral values that AI should be aligned with. The encyclical did not explicitly mention [...] ---Outline:(00:36) Pope Leo XIV Publishes Encyclical on AI(02:58) How AI Betrayal Could Deter Reckless AI Use(07:01) AI Solves Well-Known Open Mathematics Problem(09:29) In Other News(09:32) Government(10:30) Industry(11:12) Civil Society --- First published: June 3rd, 2026 Source: <a href="https://newsletter.safe.ai/p/aisn-74-the-popes-encyclical-and?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Source+URL+in+episode+description&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">https://newsletter.safe.ai/p/aisn-74-the-popes-encyclical-and</a> --- Want more? Check out our <a href="https://newsletter.mlsafety.org/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Episode+description+footer" target="_blank" rel="noreferrer">ML Safety Newsletter</a> for technical safety research. Narrated by <a href="https://type3.audio/?utm_source=TYPE_III_AUDIO&utm_medium=Podcast&utm_content=Narrated+by+TYPE+III+AUDIO&utm_term=center_for_ai_safety&utm_campaign=ai_narration" rel="noopener noreferrer" target="_blank">TYPE III AUDIO</a>. ---<div style="max-width: 100%";>Images from the article:<a href="https://substackcdn.com/image/fetch/$s_!FxvK!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ffe6ee6ff-728d-48a4-9ec5-60900d514853_600x338.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!FxvK!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ffe6ee6ff-728d-48a4-9ec5-60900d514853_600x338.png" alt="Religious figure in red vestments signing document titled "Magnifica Humanitas"" style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!cthN!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb21d25db-c845-4e6f-881d-f6afd7c18486_1054x1362.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!cthN!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb21d25db-c845-4e6f-881d-f6afd7c18486_1054x1362.png" alt="Book cover titled "AI Deterrence by Betrayal" with abstract network pattern." style="max-width: 100%;" /></a><hr style="margin-top: 24px; margin-bottom: 24px;" /><a href="https://substackcdn.com/image/fetch/$s_!8duC!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F5b73fd06-caee-432c-9016-da965d0ab56c_2012x2048.png" target="_blank"><img src="https://substackcdn.com/image/fetch/$s_!8duC!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F5b73fd06-caee-432c-9016-da965d0ab56c_2012x2048.png" alt="One novel construction from the new solution to the unit distance problem." style="max-width: 100%;" /></a>Apple Podcasts and Spotify do not show images in the episode description. Try <a href="https://pocketcasts.com/" target="_blank" rel="noreferrer">Pocket Casts</a>, or another podcast app.</div>