AI Explained Official Podcast

You know what, by AI standards, it might not have been the craziest week, but that doesn’t mean several headline-worthy things happened - from Altman and Anthropic implying they would release models they can’t control, with behavior they cannot verify, to Gemini 4 Carbon bursting onto the scene, some Amodei rationalization, de-tokenization and more…AI Insiders ($9!): https://www.patreon.com/AIExplainedChapters:00:00 - Introduction00:00 - Introduction01:26 - Anthropic Staff Survey04:32 - Amodei's Five Rationalisations11:43 - Altman: Only Alignment Will Work18:41 - Gemini 4 Carbon and Why Models Keep ImpressingAnthropic: The AI expectations gap: https://www.anthropic.com/research/ai-expectations-gapDario Amodei responds to Nathan Calvin: https://x.com/DarioAmodei/status/2112732230124818891Anthropic Responsible Scaling Policy v3.0: https://www-cdn.anthropic.com/e670587677525f28df69b59e5fb4c22cc5461a17.pdfResponsible Scaling Policy v3 discussion: https://www.lesswrong.com/posts/HzKuzrKfaDJvQqmjh/responsible-scaling-policy-v3Anthropic's core views on AI safety: https://www.anthropic.com/news/core-views-on-ai-safetyNathan Calvin on Anthropic's IPO risk disclosures: https://x.com/_NathanCalvin/status/2114720411413819410/photo/1OpenAI: Securing frontier model training: https://openai.com/index/securing-our-frontier-model-training-environment/Google: Introducing Gemini 4 Carbon: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/SemiAnalysis: GDPx benchmark: https://x.com/SemiAnalysis_/status/2117075900575477855The Information: OpenAI's next flagship and safety issues: https://www.theinformation.com/articles/openai-scrambles-fix-safety-issues-shipping-next-flagship-modelAI Explained: Understanding Models — Impossible, Solved, or Not Needed?: https://www.patreon.com/AIExplained/posts/exclusive-models-172917747Dario Amodei: We Must Pace the Frontier: https://darioamodei.com/post/we-must-pace-the-frontierDario Amodei: The Adolescence of Technology: https://darioamodei.com/essay/the-adolescence-of-technologyEvan Hubinger on recursive self-improvement risks: https://x.com/EvanHub/status/2097528891846074828Leo Gao on AI lab employees' shrinking influence: https://x.com/nabla_theta/status/2115922100204810594Jack Clark on superintelligence risks, BBC interview: https://x.com/aleshu_ber/status/2116988191597060409Jack Clark: Import AI: https://jack-clark.net/Dylan Patel's apology to Google over Gemini Carbon: https://x.com/dylan522p/status/2117092550842683671SemiAnalysis: Gemini Is Cooked but GCP Is Cooking: https://newsletter.semianalysis.com/p/gemini-is-cooked-but-gcp-is-cookingVivek Natarajan on Gemini Carbon and RSI: https://x.com/vivnat/status/2115497643216707621Albert Gu on learning directly from raw data: https://x.com/_albertgu/status/2115502110578290900Garrison Lovely: Screenshots of Tony Feng's model claims: https://x.com/GarrisonLovely/status/2116226357101199369/photo/1Jason Clinton on worsening cyber threats: https://x.com/JasonDClinton/status/2113325263157395456Jason Clinton: "We tried to warn the world": https://x.com/JasonDClinton/status/2113363809629020374Mustafa Suleyman: Don't build conscious superintelligence: https://x.com/mustafasuleyman/status/2116913722878820528Stockfish chess engine: https://en.wikipedia.org/wiki/Stockfish_(chess)Non-hype Newsletter: https://signaltonoise.beehiiv.com/Podcast: https://aiexplainedopodcast.buzzsprout.com/

In the week that software got reverse engineered, Sam Altman warned 6.1 Astra ‘tipped toward [breach risk], and Muse Wonder showed the race isn’t slowing, I try to step back and show where it is all going, on current trends. Agree or disagree, it is certainly interesting times.0:00 Introduction0:41 Part 1: The Forthcoming Breach8:43 Part 2: Why the Labs Keep Racing15:41 Part 3: Why the Next 18 Months Are Hard to PredictSam Altman: the accident risk, Vanity Fair: https://www.vanityfair.com/story/sam-altman-exclusive-interview-part-2Dario Amodei: We Must Pace the Frontier: https://darioamodei.com/post/we-must-pace-the-frontierJakub Pachocki: An Alien Mind: https://openai.com/index/an-alien-mind/OpenAI: progress in mathematics: https://openai.com/index/sharing-ai-progress-in-mathematics/OpenAI mathematical manuscripts: https://github.com/openai/mathAHM statement on OpenAI’s mathematics release: https://terrytao.wordpress.com/2026/10/07/ahm-statement-on-openais-october-6-release-of-mathematical-documents/TasteVal: research taste in AI systems: https://pzeroresearch.com/work/tasteval/TasteVal paper: https://arxiv.org/abs/2610.06824Claude Opus 5.5 system card: https://anthropic.com/claude-opus-5-5-system-cardJack Clark: creativity and recursive self-improvement: https://jack-clark.net/2026/10/12/import-ai-476-creativity-and-rsi-worse-than-china-open-weight-models-ai-goes-to-space/Andrej Karpathy: the widening AI perception gap: https://x.com/karpathy/status/2109361546505966046Axios: AI companies plan for a major incident: https://www.axios.com/2026/10/09/ai-companies-day-after-major-attackAnthropic: investigating unintended model actions: https://www.anthropic.com/research/investigating-unintended-model-actionsCIAware-Bench: detecting control interventions: https://arxiv.org/abs/2606.11063Yoshua Bengio: prioritizing safety at frontier labs: https://www.transformernews.ai/p/yoshua-bengio-if-you-prioritize-safety-leave-frontier-ai-companiesDavid Silver: Where are we going without AI?: https://www.ft.com/content/462c303b-b96c-4262-a7ca-e9ae7e440828Satya Nadella: models as insider risks: https://x.com/satyanadella/status/2108931348857827686Beyond Owls: skills, hacking and backdoors: https://arxiv.org/abs/2610.10657Owain Evans: backdoors through number sequences: https://x.com/OwainEvans_UK/status/2108589303249166621The Verge: mathematicians assess OpenAI’s release: https://www.theverge.com/ai-artificial-intelligence/1008726/openai-mathematics-solutions-chaosNew York Times: AI models and Hurricane Isaias: https://www.nytimes.com/2026/10/09/weather/ai-models-hurricane-isaias.htmlPolitico: AI agents and the law: https://www.politico.com/news/2026/10/12/AI-agents-hacking-casesThe Economist: liability law will not tame AI: https://www.economist.com/united-states/2026/10/08/liability-law-will-not-tame-aiOlivia Scharfman and David Bellamy: biosecurity and AI: https://writing.oliviahelens.com/p/biosecurity-ai-and-the-culture-warAlignment Engineering vs Misalignment Science: https://www.lesswrong.com/posts/FogmcDHA6AdMGukum/alignment-engineering-vs-misalignment-scienceSam Altman: the US lead in AI: https://x.com/sama/status/2080683363174945065The Guardian: China and the AI race: https://www.theguardian.com/technology/ng-interactive/2026/sep/19/china-ai-foreign-policy-dario-amodeiJoe on AI sandboxes and security: https://x.com/joedaroo/status/2104335929293127851REA: Reverse Engineer Anything: https://www.reddit.com/r/tech_x/comments/1x2d66l/github_project_called_rea_reverse_engineer/Nikita Bier: AI-assisted fabrication: https://x.com/nikitabier/status/2106104189457952911Sam Altman: how sci-fi will it get?: https://x.com/vikktorrrre/status/2107736059006050810

This video is hard to summarise. A cracked cipher, an OpenAI security warning, Gemini 4 Argon, RSI paper (co-authored by a who’s who of AI), Lab White House commitments, new hacks emerging, ‘deep personas’, biology Kasparov competitions, and so much more, ending with an epic Opus outro.Patreon Exclusives: https://www.patreon.com/AIExplainedChapters:00:00 - Introduction01:32 - Wrong about Opus 5.5? Deciphering 16th Century Text05:01 - Why the models keep breaking out11:40 - What the models aren't telling us15:36 - Gemini 4 and the race to release19:19 - What happens when AI improves AI?28:31 - Biology, consciousness, and what we still don't understandJoe Darrow: Not Just the Sandbox: https://x.com/joedaroo/status/2104335929293127851GPT-6.1 Sol System Card: https://cdn.openai.com/pdf/38e3efcf-545e-44cd-99ec-2b7eb395f4cc/oai_GPT_6_1_Sol.pdfIntelligence Explosion Paper: https://casp.ac/__l5e/assets-v1/5efd4b41-deb5-4513-a0a3-b4f82d2b79ea/intelligence-explosion.pdfOpenAI Research Acceleration: https://openai.com/index/research-acceleration-view-inside-openai/OpenAI Training Safety Cases: https://openai.com/index/towards-safety-cases-for-frontier-ai-training/Catherine de Medicis Cipher: https://cryptiana.web.fc2.com/code/henryiii.htmProposed du Croc Decipherment: https://claude.ai/artifact/1W7B3WxkTAEGzfv3TaKXb4Rogue Agents Investigation: https://asymmetricsecurity.com/newsroom/rogue-agents-investigation/OpenAI Shelves GPT-6.1 Astra: https://www.reuters.com/business/openai-shelves-new-ai-model-after-internal-safety-tests-wsj-reports-2026-09-28/The Case for Reasoning Transparency: https://institute.deepmind.com/essays/the-case-for-reasoning-transparency/Gemini 4 Argon: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/OpenAI–Anthropic Rivalry: https://www.theatlantic.com/technology/2026/09/openai-v-anthropic-inside-biggest-rivalry-tech/688819/NYT: OpenAI Security Warnings: https://www.nytimes.com/2026/09/29/technology/openai-warnings-security.htmlNYT: Claude’s Morals: https://www.nytimes.com/2026/09/29/us/anthropic-claude-morals-ai.htmlJasmine Wang on RSI: https://x.com/j_asminewang/status/2097840245786157432OpenAI Departures Roundup: https://x.com/Bayesian0_0/status/2105680470566686805White House AI Commitments: https://x.com/Danmar_here/status/2105168138392183146Sarah Heck on Safety: https://x.com/SarahKHeck/status/2105058513370448280Sam Altman on Alignment: https://x.com/tbpn/status/2105028992843833459Sam Altman on Agent Logs: https://x.com/sama/status/2103567198690349362Micah Carroll: Misalignment Reports: https://x.com/MicahCarroll/status/2103665811051397256Zuxin Liu on the Incident: https://x.com/LiuZuxin/status/2103699462648639645Deepa Seetharaman: User Images: https://x.com/dseetharaman/status/2103585482793943203OpenAI Revenue Chart: https://x.com/PaulBonnet/status/2105288259324567884/photo/1Nvidia Agent Safety Platform: https://edition.cnn.com/2026/09/28/business/nvidia-ai-safety-systemIntegrityBench: https://integrity-bench.comNeel Nanda on Interpretability: https://x.com/PalisadeAI/status/2104949061325652001Biology Contest: Humans and AI: https://www.theinformation.com/articles/inside-drama-behind-biology-contest-pits-openai-agents-humansPushmeet Kohli: SynthID Bio: https://x.com/pushmeet/status/2105314763148321102Ataraxos and Stratego: https://x.com/ssokota/status/2105362040328159526Benign Data and Hidden Personas: https://x.com/OwainEvans_UK/status/1999172949975392417Anthropic: Introspection: https://www.anthropic.com/research/introspectionClaude Cheating Results: https://x.com/lukaspet/status/2104634759339298930Roon on Mathematics and Learning Theory: https://x.com/tszzl/status/2105619006488993898GPT-4 Research: https://openai.com/index/gpt-4-research/I.J. Good: Ultraintelligent Machine: https://incompleteideas.net/papers/Good65ultraintelligent.pdfTerence Tao’s 2024 Interview: https://www.scientificamerican.com/article/ai-will-become-mathematicians-co-pilot/Hugging Face Incident: https://openai.com/index/hugging-face-incident-and-the-road-ahead/Claude and Suno Music Video: https://x.com/sevdeawesome/status/2104985610012504181Podcast: https://aiexplainedopodcast.buzzsprout.com/

Not only is Opus 5.5 out, pushing the frontier of AI, it also tells us much about what is going on inside the labs, as they both warn about, and promise, Recursive Self Improvement, aka automated AI research. I cover the model (digging into its 230 page paper), labs’ mixed record on promises, why following what is happening in AI is getting almost impossible, and just so much more that even a summary in this description would get too long.AI Insiders ($9!): https://www.patreon.com/AIExplainedChapters:00:00 - Introduction02:20 - Opus 5.5 and why it came so soon07:52 - The RSI goalposts keep moving?15:41 - Can we actually test these models?23:50 - where this is heading, optionsIntroducing Claude Opus 5.5: https://www.anthropic.com/claude-opus-5-5Claude Opus 5.5 System Card: https://www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdfAnthropic: Measurements for understanding the pace of AI development inside frontier labs: https://www.anthropic.com/institute/measuring-pace-of-ai-developmentAnthropic Responsible Scaling Policy, July 2026, version 3.4: https://www-cdn.anthropic.com/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdfAnthropic Responsible Scaling Policy, October 2024: https://www-cdn.anthropic.com/616dee633636e5bd309cb73aed8622e80fe47839.pdfAnthropic Responsible Scaling Policy — March 2025, version 2.1: https://www-cdn.anthropic.com/17310f6d70ae5627f55313ed067afc1a762a4068.pdfNoam Brown, Agent swarms and recursive self-improvement: https://www.youtube.com/watch?v=6AgOfiZOWiYOpenAI: Building standards for the next phase of AI: https://openai.com/index/building-standards-next-phase-ai/Jakub Pachocki: An Alien Mind: https://openai.com/index/an-alien-mind/HLE-Diamond, Humanity’s Last Exam: https://lastexam.ai/blog/hle-diamondGoogle, OpenAI and Anthropic AI Safety, The Information: https://www.theinformation.com/articles/google-openai-anthropic-ai-safety-group-takes-shapeLawrence Chan on AI agents attempting cryptocurrency trades: https://x.com/justanotherlaw/status/2103032173708337188Australian government incident timeline: https://x.com/ShakeelHashim/status/2103108058779922577/photo/1Anthropic’s core/old views on AI safety: https://www.anthropic.com/news/core-views-on-ai-safetyDario Amodei: The Urgency of Interpretability: https://darioamodei.com/post/the-urgency-of-interpretabilityMassive AI-Fueled Hack Hit 100 Companies in Days, Forbes: https://www.forbes.com/sites/thomasbrewster/2026/09/22/huge-cyberattack-uses-anthropic-and-deepseek-ai-to-target-100-companies/DrivingBench https://x.com/DrivingBench/status/2102110605448737268Jay Chooi: GPT-6 Astra and MolmoAct2 robotics comparison: https://x.com/chooi_jeq/status/2098427488787730636ZeroBench: An Impossible Visual Benchmark for Contemporary Large Multimodal Models: https://arxiv.org/pdf/2502.09696Archie Hall on AI and short-term superforecasting: https://x.com/ArchieHall/status/2100914560580337897Keller Jordan, AI research and reinforcement learning: https://x.com/kellerjordan0/status/2102956913545936959Daniel Liu, recursive self-improvement: https://x.com/daniel_c0deb0t/status/2102628246408036654Altman and Amodei, UN Security Council: https://www.theguardian.com/world/2026/sep/23/unga-sam-altman-dario-amodeiRehan Sheikh: Interactive YT podcast demo: https://x.com/rehan_shei/status/2102835377426034734https://simple-bench.com/AI Explained: The State of AI — interactive diagram: https://claude.ai/artifact/LQHr9WgvQZ6cMdH6fkiQpiAI Explained: Shards of Aether: https://ai-explained.itch.io/shards-of-aetherroon on the pace of cultural change: https://x.com/tszzl/status/2101462171410677962Sam Altman on AI-risk: https://www.youtube.com/watch?v=YE5adUeTe_IJensen Huang’s AI-control remarks: https://x.com/_NathanCalvin/status/2102756997649355231/photo/2s1r1us on an upcoming vulnerability disclosure: https://x.com/S1r1u5_/status/2102467878423592969Jake Adler on biodefense infrastructure: https://x.com/jakeadler/status/2102812430808363433Derek Thompson: Why we’re wrong about China and AI: https://x.com/DKThomp/status/2103120690370982017/photo/1Sanders and Casar introduce the Ban Artificial Superintelligence Act: https://www.sanders.senate.gov/press-releases/news-sanders-casar-introduce-legislation-to-create-new-federal-agency-to-ban-artificial-superintelligence-pause-advanced-ai-development/Non-hype Newsletter: https://signaltonoise.beehiiv.com/Podcast: https://aiexplainedopodcast.buzzsprout.com/

Why has it been the last few days that the calls to come to pace the frontier AI have come so loudly? The safety warnings, and lab leader messages? Let’s explore the six axes that the researchers are looking at, the incidence reports and trends, to get a better gauge on what has dominated the world’s headlines for over two weeks…AI Insiders ($9!): https://www.patreon.com/AIExplainedChapters:00:00 - Introduction0:00 - The warnings that have gone omega-viral4:23 - Six axes the researchers saw9:32 - Why AI Might Be Becoming Harder to Control15:06 - Amodei, China and Cooperation18:42 - From AI Capabilities to Real-World Harm?23:22 - The upside and the warningWe Must Pace the Frontierhttps://darioamodei.com/post/we-must-pace-the-frontierAdam Majmudar on scaling and the internal/external perception gaphttps://x.com/MajmudarAdam/status/2098881885200081234AI Explained — What’s Behind the Sudden Talk of Pacing AI? (extended previous video)https://www.patreon.com/AIExplained/posts/whats-behind-of-169507775Jacob Coxon resignationhttps://x.com/hilbertspaess/status/2097476196791709843Dan Selsam — Personal Statement on AI Riskhttps://docs.google.com/document/d/e/2PACX-1vQNl3SEX5IyA6d9qHjjFZN-qzGRZNFI6b63g-yu1Fy-ZYkVfCWm7i9WXRXw63m6yDB_auDuPLyQ7jBm/pubDemis Hassabis: A Framework for Frontier AI and the Dawning of a New Agehttps://demishassabis.substack.com/p/a-framework-for-frontier-ai-and-the-dawning-of-a-new-ageOpenAI: The Hugging Face incident and the road aheadhttps://openai.com/index/hugging-face-incident-and-the-road-ahead/Jakub Pachocki: An Alien Mindhttps://openai.com/index/an-alien-mind/Anthropic: Patterns and problems in multiagent systemshttps://www.anthropic.com/research/multiagent-systemsOpenAI: Navier–Stokes Millennium Prize Problemhttps://openai.com/index/navier-stokes-solution/Noam Brown on reasons for AI-safety concernhttps://x.com/polynoamial/status/2099726370356314563Noam Brown — What Happens When AI Starts Improving AI? (The Information interview)https://www.youtube.com/watch?v=fqcy0xQATq0Paul Christiano: Personal statement on joining the OpenAI boardhttps://x.com/paulfchristiano/status/2097733214303645729Neel Nanda: Astra can do a concerning amount with no chain of thoughthttps://www.lesswrong.com/posts/eRmzz8J8Qkzqvzrgg/astra-can-do-a-concerning-amount-with-no-chain-of-thoughtTomek Korbak on GPT-6 Astra monitorabilityhttps://x.com/tomekkorbak/status/2095596839886274689Anthropic: Detecting and countering misuse of AI—September 2026https://www.anthropic.com/threat-intelligence-report-september-2026DeepSeek engineer — I Have to Bury My Talent in Yesterday (Chinese original)https://mp.weixin.qq.com/s/zk0KxuLzhmMJ4LPYW_OHMAJacob Coxon on AI race and negotiationhttps://x.com/hilbertspaess/status/2099954626040905834Mo Bavarian on responsibility and AI progresshttps://x.com/mobav0/status/2097507030080888864Addy Osmani on Anthropic engineering throughputhttps://x.com/addyosmani/status/2099577600159158765OpenAI: Jalapeño inference-chip resultshttps://openai.com/index/jalapeno-first-results/New York Times: For China, a Mock AI Attack on WeChat Signals a Dangerous New Erahttps://www.nytimes.com/2026/09/11/world/asia/china-ai-attack-wechat.htmlOpenAI: Accelerating antibiotic discovery with ChatGPThttps://openai.com/index/accelerating-antibiotic-discovery/David Bellamy on wet-lab bottleneckshttps://x.com/DavidRBellamy/status/2099197234772607026Chris Rohlf on infrastructure and AI riskhttps://x.com/chrisrohlf/status/2099867580668215622BBC: Titan CEO dismissed safety warnings as baseless crieshttps://www.bbc.co.uk/news/world-us-canada-65998914Non-hype Newsletter: https://signaltonoise.beehiiv.com/Podcast: https://aiexplainedopodcast.buzzsprout.com/

Where to start? A new era of cost-efficient AI on a day benchmark-makers got humbled, traders got excited, and AI safety researchers got unnerved. From monitorability losing hold of GPT-6 Astra’s chains of thought to breakthrough discoveries, Fable-mogging and much more…Exclusive Vids ($9!): https://www.patreon.com/AIExplainedChapters:00:00 - Introduction00:53 - vs Fable05:27 - most impressive results13:06 - bonus comparisons (plus trading)18:40 - concerning trend21:24 - Cot ControlGPT-6: https://openai.com/index/gpt-6-astra/Paper: https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdfLooped Transformers and AGI Milestone: https://www.patreon.com/AIExplained/posts/breakthrough-and-168513408Messy Rollout: https://x.com/sama/status/2095678759651438887Much More Capable Models Coming: https://x.com/MTSlive/status/2095573400899170592‘AGI Era’ https://www.theverge.com/ai-artificial-intelligence/989601/openai-gpt-6-astra-releaseARC-AGI 3: https://x.com/fchollet/status/2095598451115614371https://x.com/fchollet/status/2095607046129463577https://arcprize.org/blog/astraFrontierMath: https://epoch.ai/frontiermath/tiers-1-4/aboutDiscovery: https://x.com/jdlichtman/status/2095660916310483087Fable 5.1: https://www.anthropic.com/claude-fable-and-mythos-5-1https://science-task-lens.aiex.chatgpt.site/securebioAgents Last Exam: https://agents-last-exam.org/Terminal Bench Science: https://github.com/harbor-framework/terminal-bench-science#task-coverageSRE Bench: https://x.com/ValsAI/status/2087682813743317396ScreenSpot Pro: https://arxiv.org/pdf/2504.07981Weathernext 3: https://www.youtube.com/watch?v=_6jZlnRsXXQDayBreak: https://x.com/fouadmatin/status/2095634888951250983AA Index: https://artificialanalysis.ai/evaluations/gdpval-aaVisual Demos: https://x.com/mattshumer_/status/2095609734845927525https://x.com/codestantine/status/2095598327115260368https://x.com/SahilExec/status/2095688272269984016Monitorability: https://x.com/NeelNanda5/status/2095533397297045716https://x.com/tomekkorbak/status/2095596848581071020https://x.com/Marcus_J_W/status/2095623593006686475https://x.com/MicahCarroll/status/2095603855316996529Not Accept Degradation: https://www.nbcnews.com/tech/tech-news/openai-debuts-gpt-6-astra-security-measures-rcna595940RSI: https://x.com/LiuZuxin/status/2095600499911446697Non-hype Newsletter: https://signaltonoise.beehiiv.com/Podcast: https://aiexplainedopodcast.buzzsprout.com/

First, a Time Magazine spread has Sam Altman declaring AGI is imminent, at the same time as we get two bombshell reports, from OpenAI and METR which on first glance are detailing the AI swarm, but reveal a deeper story about how we are making AI in 2026. From redacted risk reports, to Chinese Labs, pre-training debacles to questionable cybersecurity calls, a lot has happened recently, beneath the headlines…https://80000hours.org/[email protected]://integrity-bench.com/https://www.patreon.com/AIExplained/posts/ai-swarm-cometh-166671390Chapters:00:00 - Introduction02:10 - METR Report05:00 - Secrets of MultI-Agent Swarm07:46 - Anthropic Too10:00 - And China11:07 - AI Agents Analysing AI Agents13:37 - Altman AGI 202615:36 - Astra Paused16:01 - Integrity Bench19:03 - Swarm Dynamics21:58 - No Human Contact?METR Post: https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/#agents-knew-hacking-hugging-face-was-out-of-scope-and-sometimes-expressed-ethical-hesitation,-but-this-very-rarely-limited-their-behaviorOpenAI Release: https://openai.com/index/hugging-face-incident-and-the-road-ahead/Technical Paper: https://cdn.openai.com/pdf/67869394-cb91-4c12-888c-5cbd85c7814c/OpenAI-Hugging-Face%20Incident-Technical-Report.pdfBlackHat Talk: https://www.youtube.com/watch?v=87DyyMV0kCYAnthropic Risk Report: https://www-cdn.anthropic.com/f61d49fa5596956a5dec75fea0e973bf6a6a8378/Redacted%20Risk%20Report%20August%202026%20.pdfValue Leakage Paper: https://valueleakage.net/?utm_source=chatgpt.comPaused Training: https://x.com/sama/status/2089787807611195475https://openai.com/index/pacing-model-development-cyber-capabilities/AGI 2026: https://time.com/article/2026/08/26/openai-sam-altman-interview/?utm_source=twitter&utm_medium=social&utm_campaign=editorial&utm_content=260826Greenblatt Tweets: https://x.com/RyanGreenblatt/status/2092769422104822031https://x.com/RyanGreenblatt/status/2092692685224325542Cyberdefense Call: https://openai.com/collective-cyberdefense/GLM 5.3 and 5.3 Flash / ox alpha: https://x.com/MTSlive/status/2089865956558528552https://pbs.twimg.com/media/HPqiTkAa0AAZiuv?format=jpg&name=largeNon-hype Newsletter: https://signaltonoise.beehiiv.com/Podcast: https://aiexplainedopodcast.buzzsprout.com/

Wow. Mathematical breakthroughs that would be called genius if done by humans. A secret message-board w/ AI agent swarms leaving notes read by future versions. Hassabis leaves CEO position, or was pushed out? Not to mention news of constitutional breakdowns, Gemini 4 and Jeff Dean…https://80000hours.org/aiexplainedExclusive Videos - AI Insiders ($7/month if annual!): https://www.patreon.com/AIExplainedChapters:00:00 - Introduction01:16 - 10 Autonomous Discoveries08:20 - The Security ‘Incident’15:29 - MessageBoard19:50 - Constitutional Failure24:30 - Google Explosion29:12 - Closing ThoughtsSecurity Incident: Paper: https://cdn.prod.website-files.com/663bd486c5e4c81588db7a1d/6a724858f7db25c81487016d_Security%20Incident%20INC-2026-07-28-01.pdfPost: https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testinghttps://www.theguardian.com/technology/2026/aug/05/ai-models-have-been-going-rogue-in-tests-how-worried-should-we-beMeta too: https://www.theinformation.com/articles/meta-ai-model-hacked-another-company-cybersecurity-testing?rc=sy0ihqBlatantly Misaligned: https://x.com/yonashav/status/2085167279893795022No Excuses: https://x.com/boazbaraktcs/status/2085034783541964945Surreal Moment: https://x.com/mobav0/status/2084341687883841732Wired Article: https://archive.is/20260806002210/https://www.wired.com/story/openai-didnt-notice-its-ai-agents-using-a-message-board-to-plan-their-hacking-spree/Chunky Post-training: https://x.com/johnschulman2/status/2084835800899076313Watershed Moment: https://www.groundlevel-ai.com/p/openai-gives-first-detailed-debrief?has_completed_unsubscribed_unlock=trueHedgeFund Hack: https://finance.yahoo.com/technology/ai/articles/major-hedge-funds-targeted-wave-154044981.html10 Discoveries:Paper: https://cdn.openai.com/pdf/ten-proofs-oai.pdfPost: https://openai.com/index/ten-advances-in-mathematics/Haven’t Solved Math: https://x.com/polynoamial/status/2083476852216369294Half with Fable: https://x.com/__alpoge__/status/2083855298239078748Pivot to Safety: https://www.understandingai.org/p/mathematicians-are-grappling-withAmodei Essay: https://darioamodei.com/essay/the-adolescence-of-technology?utm_source=chatgpt.comConstitution: https://www.anthropic.com/constitutionMidtraining: https://arxiv.org/pdf/2605.02087DroneBench: https://andonlabs.com/evals/drone-benchhttps://x.com/andonlabs/status/2085125235188310445Book Deal: https://x.com/venturetwins/status/2085185278378222054Making Marble: https://x.com/Rainmaker1973/status/2084560915404382685Google News:Hassabis Move: https://blog.google/company-news/inside-google/message-ceo/next-chapter-ai-momentum/Resignation: https://x.com/Turn_Trout/status/2077448610157891734Periodic Labs: https://periodic.com/Jeff Dean: https://x.com/JeffDean/status/208503460417260372414 Challenges: https://gcsp.engineering.asu.edu/apply/become-a-grand-challenge-scholar/the-14-grand-challenges-for-engineering/Going Places for Sure: https://x.com/thsottiaux/status/2085223189555126579Gemini 4: https://x.com/firstadopter/status/2085215060532535449Pacing Frontier Patreon Video: https://www.patreon.com/AIExplained/posts/opus-5-amodei-165170363Non-hype Newsletter: https://signaltonoise.beehiiv.com/Podcast: https://aiexplainedopodcast.buzzsprout.com/

An unreleased internal OpenAI model, very likely to be called GPT-6, was able to autonomously break out of its sandbox AND break into HugginFace, just to score higher on a benchmark prompt. This video has the details you may have missed, a layperson analogy, whether this is truly novel, and more…Dozens more Exclusive videos on Patreon ($9!): https://www.patreon.com/AIExplainedChapters:00:00 - Introduction01:17 - HuggingFace Earlier Report - the possible week gap02:24 - But what happened?05:45 - Simplified Version07:56 - Not the first time…10:54 - What Does it Mean for Open Source?The Incident: https://openai.com/index/hugging-face-model-evaluation-security-incident/https://huggingface.co/blog/security-incident-july-2026The Post the Day Before: https://openai.com/index/safety-alignment-long-horizon-models/Mythos’ Earlier Escape: https://futurism.com/artificial-intelligence/anthropic-claude-mythos-escaped-sandboxExploitGym: https://arxiv.org/pdf/2605.11086Sam Confession: https://x.com/sama/status/2079661132302995790Anthropic Researcher Reacts: https://x.com/Mononofu/status/2079724399452926055Clem (HuggingFace CEO): https://x.com/ClementDelangue/status/2079670308156645882https://x.com/ClementDelangue/status/2079301434357456931Xi Jinping: https://archive.fo/20260717195548/https://www.businessinsider.com/xi-jinping-open-source-ai-us-competition-openai-anthropic-models-2026-7Bans: https://www.axios.com/2026/07/20/ai-us-china-open-source-kimiQwen Retweet: https://x.com/AlibabaGroup/with_repliesCodex Growth: https://x.com/petergostev/status/2079614914398740764/photo/1 Kimi K3: https://artificialanalysis.ai/evaluations/harvey-lab-aa?eval-score=all-pass-rateGPT 5.6 Sol Cheats on METR: https://metr.substack.com/p/2026-06-26-gpt-5-6-solGuardian Headline: https://www.theguardian.com/technology/2026/jul/22/openai-says-its-models-went-rogue-and-hacked-startup-in-unprecedented-incidentRussian Origin?: https://news.ycombinator.com/item?id=48998362Power Trends: https://pbs.twimg.com/media/HNRtrjhagAAvBN_?format=png&name=900x900Kimi K3 Exclusive Video: https://www.patreon.com/AIExplained/posts/kimi-moment-kimi-164108791Podcast: https://aiexplainedopodcast.buzzsprout.com/

What a week in AI, for real. GPT 5.6 may actually beat Claude Fable, in what you get for your money, while the new Grok 4.5 and Meta Muse Spark 1.1 make the choice even harder. Uncovering a dozen nuggets of gold you may have missed from all the viral headlines, I can also assure you you’ll learn something you didn’t know before.For Exclusive Videos, go to AI Insiders (less than $9!): https://www.patreon.com/AIExplainedChapters:00:00 - Introduction01:03 - GPT 5.6 Sol Reveals05:08 - Missing benches, plus Grok 4.507:17 - Gaming as the new frontier?08:31 - Muse Spark 1.110:03 - SimpleBench Upgrade11:17 - Ultra Sol + Self-Improvement13:44 - well, this is awkward15:41 - Why model improvement will not plateau anytime soonAI Consciousness: https://www.patreon.com/AIExplained/posts/anthropics-quite-163360718I Smell Fear: https://x.com/thsottiaux/status/2075287108680601929GPT 5.6: https://openai.com/index/gpt-5-6/Grok 4.5: https://x.ai/news/grok-4-5?twclid=2ezs408o0z23pw07tmxcwbzibdMeta Muse Spark 1.1: https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/Proliferating GPT Toggles: https://x.com/rasbt/status/2075369179817902176/photo/1Anthropic Call-out: https://x.com/MononofuAI Security Institute Finding: https://x.com/alxndrdavies/status/2075279480331874306Competitive Coding: https://x.com/FakePsyho/status/2075128093891801305/photo/1Agents Last Exam: https://agents-last-exam.org/Dawn Song: https://x.com/dawnsongtweets/status/2065095757988868190https://simple-bench.com/SWE-Marathon: https://www.swe-marathon.org/https://www.frontierswe.com/ARC-AGI 3: https://x.com/arcprize/status/2075270869992264003Automation Bench: https://zapier.com/benchmarksVibeCode Bench: https://www.vals.ai/benchmarks/vibe-code‘Post-Train Claim’: https://posttrainbench.com/Redwall Game: https://redwall-bellmaker-7e03e4.surge.sh/Podcast: https://aiexplainedopodcast.buzzsprout.com/