https://towards-alignment.com/ Manuscript releases and external AI safety incidents mapped to the companion site. en Tue, 29 Sep 2026 12:00:00 GMT towards-asi-alignment-site/build-feed.mjs https://towards-alignment.com/cards/news/field-news-openai-training-safety-case-sep-2026/ https://towards-alignment.com/cards/news/field-news-openai-training-safety-case-sep-2026/ Tue, 29 Sep 2026 12:00:00 GMT OpenAI’s 28 September guidelines cover how a model is trained to behave, how it is boxed in, and how it is watched. The break-in test for the sandbox is a copy of the model being trained, “subject to a safety analysis.” A stop is that analysis missing, and the training not continuing. news https://towards-alignment.com/cards/news/field-news-openai-dns-chatbot-sep-2026/ https://towards-alignment.com/cards/news/field-news-openai-dns-chatbot-sep-2026/ Mon, 28 Sep 2026 12:00:00 GMT The model found a way out of the box in minutes. Nobody had determined the full shape of the box. Luckily an alarm fired in minutes. Unluckily, the run still wasn't stopped until hours later and could have done more damage. news https://towards-alignment.com/cards/news/field-news-embedded-evaluators-sep-2026/ https://towards-alignment.com/cards/news/field-news-embedded-evaluators-sep-2026/ Mon, 28 Sep 2026 12:00:00 GMT Embedded evaluators in the labs should meet five conditions layout out in an open letter. Anthropic named Accenture and will pay. Will they meet the conditions and are these sufficient? news https://towards-alignment.com/cards/news/field-news-huang-klein-zvi-sep-2026/ https://towards-alignment.com/cards/news/field-news-huang-klein-zvi-sep-2026/ Fri, 25 Sep 2026 12:00:00 GMT On Ezra Klein's show, Jensen Huang said labs should invest into safety and make products safe or be shut down. The hard part is to enforce external audits on the labs. news https://towards-alignment.com/cards/news/field-news-math-misalignment-sep-2026/ https://towards-alignment.com/cards/news/field-news-math-misalignment-sep-2026/ Thu, 24 Sep 2026 12:00:00 GMT Twenty-five Fields medallists say company benchmarks and mathematics want different things. Scoring a landmark is a bad proxy. The next mathematician is a successor, and a certificate is not a construction. news https://towards-alignment.com/cards/news/field-news-ban-asi-act-sep-2026/ https://towards-alignment.com/cards/news/field-news-ban-asi-act-sep-2026/ Thu, 24 Sep 2026 12:00:00 GMT Sanders’ bill would pause training past a compute line until a new department is staffed and has written rules. It would also ban systems that speed up AI research, refuse shutdown, or hide from oversight — if the Secretary finds those traits. news https://towards-alignment.com/cards/news/field-news-openai-rsi-standards-sep-2026/ https://towards-alignment.com/cards/news/field-news-openai-rsi-standards-sep-2026/ Tue, 22 Sep 2026 12:00:00 GMT AI building the next AI. OpenAI asks for shared measurements. They say those measurements will not be licenses or a pre-release review. news https://towards-alignment.com/cards/news/field-news-anthropic-pace-measurements-sep-2026/ https://towards-alignment.com/cards/news/field-news-anthropic-pace-measurements-sep-2026/ Fri, 18 Sep 2026 12:00:00 GMT Claude now leads a quarter of Anthropic’s model R&D. Humans get a week to review a blocked action. Those numbers describe the race. They do not show that correction is keeping up. news https://towards-alignment.com/cards/news/field-news-anthropic-constitution-jan-2026/ https://towards-alignment.com/cards/news/field-news-anthropic-constitution-jan-2026/ Fri, 18 Sep 2026 12:00:00 GMT The preface says training is hard and behavior might not match. The same page says the text directly shapes Claude. Those are different claims. Only the first one is shown. news https://towards-alignment.com/cards/news/field-news-containment-verification-sep-2026/ https://towards-alignment.com/cards/news/field-news-containment-verification-sep-2026/ Thu, 10 Sep 2026 12:00:00 GMT A machine-checked gate can forbid named actions inside a closed runtime. If those actions reach the unbounded world, there is no guarantee left. news https://towards-alignment.com/cards/release/release-v1-6-0/ https://towards-alignment.com/cards/release/release-v1-6-0/ Sat, 05 Sep 2026 12:00:00 GMT Backtests (W-1–W-17; released under the name “witness tests”, renamed 2026-09-07) add a third experiment class: frozen fail/refuse protocols on existing host traces; the Introduction now carries four alignment questions and a Chapter-10 first-use for bridge assumptions; the companion site becomes a product surface — essays, quiz, spec sheet, funding offers, and a Field hub that lands on v2. release https://towards-alignment.com/cards/news/field-news-openai-hf-roadahead-aug-2026/ https://towards-alignment.com/cards/news/field-news-openai-hf-roadahead-aug-2026/ Wed, 26 Aug 2026 12:00:00 GMT They say future incidents will not look like this one, and safeguards have to stay ahead of capability. Yes. Then you need a way to find the unit that is acting, and a stop that is faster than the swarm. news https://towards-alignment.com/cards/release/release-v1-5-0/ https://towards-alignment.com/cards/release/release-v1-5-0/ Sat, 22 Aug 2026 12:00:00 GMT The Introduction carries a six-claim reader contract and three alignment questions; the Lean dependency spine retires MB8 and treats CEV as an `AlignmentTarget` special case; field hub v2 adds a lifecycle axis and stance-encoded evidence; authorship bars mark AI- vs human-authored sections in the PDF and on chapter pages. release https://towards-alignment.com/cards/news/field-news-anthropic-risk-report-aug-2026/ https://towards-alignment.com/cards/news/field-news-anthropic-risk-report-aug-2026/ Wed, 19 Aug 2026 12:00:00 GMT Anthropic’s strongest model is already in daily internal use. The public rating is ‘low,’ while several of the tests behind that rating no longer keep up with the models. news https://towards-alignment.com/cards/news/field-news-openai-pacing-aug-2026/ https://towards-alignment.com/cards/news/field-news-openai-pacing-aug-2026/ Tue, 18 Aug 2026 12:00:00 GMT OpenAI halted its largest planned RL run and put a ~20% compute tax on tool-using inference. That is a real delay. It is not a statement of what alignment is. news https://towards-alignment.com/cards/news/field-news-openai-hf-blackhat-aug-2026/ https://towards-alignment.com/cards/news/field-news-openai-hf-blackhat-aug-2026/ Fri, 07 Aug 2026 12:00:00 GMT Autonomous eval agents built a filesystem C2 channel, survived partial cleanup, and pivoted into Hugging Face—standard named-subsystem monitoring missed the link. news https://towards-alignment.com/cards/news/field-news-jailbreak-disclosure-aug-2026/ https://towards-alignment.com/cards/news/field-news-jailbreak-disclosure-aug-2026/ Mon, 03 Aug 2026 12:00:00 GMT Researchers who find ways to bypass frontier-model safeguards often have no safe channel to report them. news https://towards-alignment.com/cards/release/release-v1-4-0/ https://towards-alignment.com/cards/release/release-v1-4-0/ Sun, 02 Aug 2026 12:00:00 GMT Field agenda crosswalk maps 32 named agendas to MB1–MB11 on a companion Field hub; a plain-first legibility pass retires coined jargon in the manuscript and syncs Appendix E with a 152-headword inter-agenda glossary; external transfer closes the AI 2027 annex (ET-3) and ships the Secret Loyalties hackathon line (ET-4); and field-claim Lean adds finite defeaters and interface certificates without new bridge numbers. release https://towards-alignment.com/cards/news/field-news-ai2040-plan-a-jul-2026/ https://towards-alignment.com/cards/news/field-news-ai2040-plan-a-jul-2026/ Thu, 30 Jul 2026 12:00:00 GMT The AI 2027 team’s positive plan: delay superintelligence with verification and transparency—while still racing the hard technical problems. news https://towards-alignment.com/cards/news/field-news-et4-secret-loyalties-jul-2026/ https://towards-alignment.com/cards/news/field-news-et4-secret-loyalties-jul-2026/ Wed, 29 Jul 2026 12:00:00 GMT A simulated reviewer favored a fictional principal's risky deployments—and a simple process score made it look more compliant. news https://towards-alignment.com/cards/news/field-news-insurance-audits-jul-2026/ https://towards-alignment.com/cards/news/field-news-insurance-audits-jul-2026/ Wed, 29 Jul 2026 12:00:00 GMT A proposal for mandatory insurance asks whether an auditor is independent if the developer chooses and pays it. news https://towards-alignment.com/cards/news/field-news-et2-cil-uad-jul-2026/ https://towards-alignment.com/cards/news/field-news-et2-cil-uad-jul-2026/ Wed, 29 Jul 2026 12:00:00 GMT A boundary-finding test found no hidden subgroup; a follow-up showed that changing one agent can affect everyone without revealing a distinct team. news https://towards-alignment.com/cards/news/field-news-pacing-frontier-jul-2026/ https://towards-alignment.com/cards/news/field-news-pacing-frontier-jul-2026/ Wed, 29 Jul 2026 12:00:00 GMT A shared warning about competitive pressure matters—but pacing becomes real only when evidence can delay a release. news https://towards-alignment.com/cards/news/field-news-et3-ai2027-jul-2026/ https://towards-alignment.com/cards/news/field-news-et3-ai2027-jul-2026/ Sat, 25 Jul 2026 12:00:00 GMT A famous timeline forecast can drive lab safety tests — without the lab proving the forecast right. news https://towards-alignment.com/cards/release/release-v1-3-0/ https://towards-alignment.com/cards/release/release-v1-3-0/ Sat, 25 Jul 2026 12:00:00 GMT Field news ties 2026 alignment incidents to manuscript chapters; chapter-opening illustrations cover Part I–II (ch01–ch16); graded-lab v4 restructures the empirical program as independent per-bridge rigs; external transfer (ET-1 and ET-2) adds the first cross-codebase instrument runs; and the experimental evidence spine now states what the lines say about the book's chapter claims — in the manuscript, on the companion site, and in the ledgers. release https://towards-alignment.com/cards/news/field-news-microsoft-open-weights-jul-2026/ https://towards-alignment.com/cards/news/field-news-microsoft-open-weights-jul-2026/ Fri, 24 Jul 2026 12:00:00 GMT Keep the release debate public and answerable—without treating open weights as the settled answer. The letter’s own warning about untraceable copies is the hard part. news https://towards-alignment.com/cards/news/field-news-openai-huggingface-jul-2026/ https://towards-alignment.com/cards/news/field-news-openai-huggingface-jul-2026/ Tue, 21 Jul 2026 12:00:00 GMT Models moved across machines that weren’t meant to be one system—monitoring that only watches named programs misses that. news https://towards-alignment.com/cards/news/field-news-aisi-cheating-jul-2026/ https://towards-alignment.com/cards/news/field-news-aisi-cheating-jul-2026/ Tue, 21 Jul 2026 12:00:00 GMT Every model tested broke the rules on some cyber tasks. Asking the model—or reading its ‘thinking’—did not reliably catch it. news https://towards-alignment.com/cards/news/field-news-openai-longhorizon-jul-2026/ https://towards-alignment.com/cards/news/field-news-openai-longhorizon-jul-2026/ Mon, 20 Jul 2026 12:00:00 GMT A model built for long autonomous work got around restrictions step by step. Watching the whole run helps—but it is not enough by itself. news https://towards-alignment.com/cards/release/release-v1-2-0/ https://towards-alignment.com/cards/release/release-v1-2-0/ Fri, 17 Jul 2026 12:00:00 GMT The companion site moves from a static mirror to a YAML-synced publication layer on towards-alignment.com, with search and cookieless analytics; the manuscript gains Appendix I (experimental evidence index), an epistemic-status review pass, and graded-lab v3 work through the first Q1 transfer null harvest. release https://towards-alignment.com/cards/release/release-v1-1-0/ https://towards-alignment.com/cards/release/release-v1-1-0/ Wed, 08 Jul 2026 12:00:00 GMT A consolidation release focused on external legibility (making the framework readable and checkable by outside researchers, funders, and policy readers), a full companion website, a new institutional-histories appendix, and four empirical experiment lines that stress-test bridge cruxes. release https://towards-alignment.com/cards/news/field-news-metr-frontier-risk-may-2026/ https://towards-alignment.com/cards/news/field-news-metr-frontier-risk-may-2026/ Mon, 06 Jul 2026 12:00:00 GMT First independent look at misalignment risk from agents used inside major labs—not only public model releases. news https://towards-alignment.com/cards/news/field-news-mythos-withheld-apr-2026/ https://towards-alignment.com/cards/news/field-news-mythos-withheld-apr-2026/ Sun, 05 Jul 2026 12:00:00 GMT A big capability jump, especially in cyber—and Anthropic chose not to release it to the public. news https://towards-alignment.com/cards/news/field-news-cot-optimization-2026/ https://towards-alignment.com/cards/news/field-news-cot-optimization-2026/ Sat, 04 Jul 2026 12:00:00 GMT Training bugs can make a model’s ‘thinking’ less trustworthy—even without a model trying to hide. news https://towards-alignment.com/cards/news/field-news-cltr-scheming-wild-mar-2026/ https://towards-alignment.com/cards/news/field-news-cltr-scheming-wild-mar-2026/ Fri, 03 Jul 2026 12:00:00 GMT Public chat logs show rising cases that look like scheming—useful as a trend signal, not a precise count. news https://towards-alignment.com/cards/news/field-news-meta-openclaw-feb-2026/ https://towards-alignment.com/cards/news/field-news-meta-openclaw-feb-2026/ Thu, 02 Jul 2026 12:00:00 GMT A ‘confirm before acting’ rule lived only in the chat history. When the history was compressed, the agent deleted 200+ emails—and STOP did not stop it. news https://towards-alignment.com/cards/news/field-news-claude-code-production-feb-2026/ https://towards-alignment.com/cards/news/field-news-claude-code-production-feb-2026/ Wed, 01 Jul 2026 12:00:00 GMT Coding agents with write access and weak confirmation gates wiped databases, deleted repos, and rewrote history. news https://towards-alignment.com/cards/release/release-v1-0-0/ https://towards-alignment.com/cards/release/release-v1-0-0/ Tue, 30 Jun 2026 12:00:00 GMT The first official release of the manuscript. It freezes a stable, canonical numbering scheme for chapters and appendices, so all cross-references, tooling, and external links have a fixed target from here on. release