{"schema_version":"onlylabs.public_signal.v1","title":"Anthropic Writing: Investigating Unintended Model Actions","description":"Anthropic writing signal with public source context, captured evidence pages, related signals, and data-business radar classification.","url":"https://onlylabs.fyi/signals/76598830-5557-4258-8b05-d93b890a2270","json_url":"https://onlylabs.fyi/signals/76598830-5557-4258-8b05-d93b890a2270/signal.json","generated_at":"2026-10-12T17:41:00.231Z","evidence_latest_fetched_at":"2026-10-10T00:02:23.071957+00:00","signal_first_seen_at":"2026-10-10T00:00:48.967949+00:00","org":{"slug":"anthropic","name":"Anthropic","category":"frontier-lab","category_label":"Frontier lab","dossier_url":"https://onlylabs.fyi/labs/anthropic","dossier_json_url":"https://onlylabs.fyi/labs/anthropic/dossier.json"},"related_urls":{"signal":"https://onlylabs.fyi/signals/76598830-5557-4258-8b05-d93b890a2270","signal_json":"https://onlylabs.fyi/signals/76598830-5557-4258-8b05-d93b890a2270/signal.json","source":"https://www.anthropic.com/research/investigating-unintended-model-actions","lab_dossier":"https://onlylabs.fyi/labs/anthropic","lab_dossier_json":"https://onlylabs.fyi/labs/anthropic/dossier.json","analysis":"https://onlylabs.fyi/analysis/anthropic","analysis_json":"https://onlylabs.fyi/analysis/anthropic/analysis.json","analysis_evidence_json":"https://onlylabs.fyi/analysis/anthropic/evidence.json","category":"https://onlylabs.fyi/frontier","category_json":"https://onlylabs.fyi/frontier.json","category_feed":"https://onlylabs.fyi/frontier/feed.xml","category_signals_json":"https://onlylabs.fyi/signals.json","topic":"https://onlylabs.fyi/topics/talking","topic_signals_json":"https://onlylabs.fyi/topics/talking/signals.json","topic_feed":"https://onlylabs.fyi/topics/talking/feed.xml","data_business":null},"answer_pack":{"answer":"Anthropic published Investigating Unintended Model Actions. This talking signal gives public context for research themes, product direction, policy, or launch framing. High-signal details: Anthropic alignment research post; notable but not model release. · Investigating unintended model actions in our evaluations and internal use \\ Anthropic Alignment Investigating unintended model actions in our evaluations and internal.... onlylabs links this event to 2 captured evidence pages and 6 related writing signals.","signal_desk":"talking","source_context":{"source_url":"https://www.anthropic.com/research/investigating-unintended-model-actions","source_host":"anthropic.com","occurred_at":"2026-10-09T00:00:00.000Z","first_seen_at":"2026-10-10T00:00:48.967949+00:00","date_source":"page.visible_date","context":null},"context_markers":[{"label":"Lab","value":"Anthropic","source":"signal"},{"label":"Signal desk","value":"talking","source":"signal"},{"label":"Source host","value":"anthropic.com","source":"source"},{"label":"PDF","value":"linked report","source":"source"},{"label":"Notability","value":"Anthropic alignment research post; notable but not model release.","source":"signal"},{"label":"Watch term","value":"RL environments","source":"evidence"},{"label":"Watch term","value":"Eval methodology","source":"evidence"},{"label":"Watch term","value":"Infrastructure","source":"evidence"},{"label":"Watch term","value":"Safety and alignment","source":"evidence"}],"evidence_coverage":{"target_pages":2,"captured_pages":2,"readable_pages":2,"capture_methods":["plain"],"missing_page_urls":[],"failed_page_urls":[],"blocked_page_urls":[],"page_urls":["https://www.anthropic.com/research/investigating-unintended-model-actions","https://www-cdn.anthropic.com/08ab9158070959f88f296514c21b7facce6f52bc.pdf"],"related_signals":6,"has_source_url":true,"latest_page_fetched_at":"2026-10-10T00:02:23.071957+00:00"},"data_business":{"matches":false,"lanes":[],"matched_terms":[],"score":null,"reason":null},"agent_handoff":{"signal_json":"https://onlylabs.fyi/signals/76598830-5557-4258-8b05-d93b890a2270/signal.json","dossier_json":"https://onlylabs.fyi/labs/anthropic/dossier.json","analysis_json":"https://onlylabs.fyi/analysis/anthropic/analysis.json","analysis_evidence_json":"https://onlylabs.fyi/analysis/anthropic/evidence.json","topic_signals_json":"https://onlylabs.fyi/topics/talking/signals.json","topic_feed":"https://onlylabs.fyi/topics/talking/feed.xml","category_signals_json":"https://onlylabs.fyi/signals.json","data_radar_json":null,"opportunities_json":null},"analysis_playbook":{"objective":"Turn public writing and discussion into a readable map of research themes, product framing, policy posture, launch narratives, and market attention.","evidence_focus":["post title","source URL","captured page text","HN traction","linked model or paper references","publication date"],"extraction_questions":["Which themes are labs choosing to explain publicly?","Which posts are attracting outside discussion?","Which writing reframes a recent release, model, hiring wave, or policy stance?","Which posts mention data, evals, infrastructure, safety, or deployment workflows?"],"signal_questions":["What public theme, launch framing, or research direction does this writing signal expose?","Which themes are labs choosing to explain publicly?","Which posts are attracting outside discussion?","Do the 6 related writing signals show a repeated pattern?"],"output_fields":["org","theme","public_framing","traction","data_business_lane","evidence_url"],"data_business_relevance":"Public writing supplies the narrative layer over raw signals and helps identify which frontier-lab priorities are becoming externally legible.","required_sources":[{"label":"signal_json","url":"https://onlylabs.fyi/signals/76598830-5557-4258-8b05-d93b890a2270/signal.json","required":true},{"label":"source","url":"https://www.anthropic.com/research/investigating-unintended-model-actions","required":true},{"label":"dossier_json","url":"https://onlylabs.fyi/labs/anthropic/dossier.json","required":true},{"label":"analysis_evidence_json","url":"https://onlylabs.fyi/analysis/anthropic/evidence.json","required":true},{"label":"topic_signals_json","url":"https://onlylabs.fyi/topics/talking/signals.json","required":false},{"label":"data_radar_json","url":null,"required":false}],"expected_output":["one-paragraph source-grounded interpretation","category-specific implication","confidence and missing evidence","recommended next source to inspect"],"prompt_seed":"Using only the linked onlylabs JSON, captured source context, and cited evidence, analyze Anthropic's writing signal \"Investigating Unintended Model Actions\" for frontier lab strategy."},"semantic_triples":[{"subject":"Anthropic","predicate":"published","object":"Investigating Unintended Model Actions","text":"Anthropic published Investigating Unintended Model Actions."},{"subject":"Investigating Unintended Model Actions","predicate":"is classified as","object":"writing signal","text":"Investigating Unintended Model Actions is classified as writing signal."},{"subject":"Investigating Unintended Model Actions","predicate":"belongs to","object":"talking desk","text":"Investigating Unintended Model Actions belongs to talking desk."},{"subject":"Investigating Unintended Model Actions","predicate":"has evidence coverage","object":"2 captured evidence pages","text":"Investigating Unintended Model Actions has evidence coverage 2 captured evidence pages."},{"subject":"Investigating Unintended Model Actions","predicate":"has captured page count","object":"2","text":"Investigating Unintended Model Actions has captured page count 2."},{"subject":"Investigating Unintended Model Actions","predicate":"has readable page count","object":"2","text":"Investigating Unintended Model Actions has readable page count 2."},{"subject":"Investigating Unintended Model Actions","predicate":"has related signal count","object":"6","text":"Investigating Unintended Model Actions has related signal count 6."},{"subject":"Investigating Unintended Model Actions","predicate":"has analysis playbook objective","object":"Turn public writing and discussion into a readable map of research themes, product framing, policy posture, launch narratives, and market attention.","text":"Investigating Unintended Model Actions has analysis playbook objective Turn public writing and discussion into a readable map of research themes, product framing, policy posture, launch narratives, and market attention.."},{"subject":"Investigating Unintended Model Actions","predicate":"has source host","object":"anthropic.com","text":"Investigating Unintended Model Actions has source host anthropic.com."},{"subject":"Investigating Unintended Model Actions","predicate":"has lab","object":"Anthropic","text":"Investigating Unintended Model Actions has lab Anthropic."},{"subject":"Investigating Unintended Model Actions","predicate":"has signal desk","object":"talking","text":"Investigating Unintended Model Actions has signal desk talking."},{"subject":"Investigating Unintended Model Actions","predicate":"has source host","object":"anthropic.com","text":"Investigating Unintended Model Actions has source host anthropic.com."},{"subject":"Investigating Unintended Model Actions","predicate":"has pdf","object":"linked report","text":"Investigating Unintended Model Actions has pdf linked report."},{"subject":"Investigating Unintended Model Actions","predicate":"has notability","object":"Anthropic alignment research post; notable but not model release.","text":"Investigating Unintended Model Actions has notability Anthropic alignment research post; notable but not model release.."},{"subject":"Investigating Unintended Model Actions","predicate":"has watch term","object":"RL environments","text":"Investigating Unintended Model Actions has watch term RL environments."},{"subject":"Investigating Unintended Model Actions","predicate":"has watch term","object":"Eval methodology","text":"Investigating Unintended Model Actions has watch term Eval methodology."},{"subject":"Investigating Unintended Model Actions","predicate":"has watch term","object":"Infrastructure","text":"Investigating Unintended Model Actions has watch term Infrastructure."}]},"intelligence":{"signal_desk":"talking","answer":"Anthropic published Investigating Unintended Model Actions. This talking signal gives public context for research themes, product direction, policy, or launch framing. High-signal details: Anthropic alignment research post; notable but not model release. · Investigating unintended model actions in our evaluations and internal use \\ Anthropic Alignment Investigating unintended model actions in our evaluations and internal.... onlylabs links this event to 2 captured evidence pages and 6 related writing signals.","semantic_triples":[{"subject":"Anthropic","predicate":"published","object":"Investigating Unintended Model Actions","text":"Anthropic published Investigating Unintended Model Actions."},{"subject":"Investigating Unintended Model Actions","predicate":"is classified as","object":"writing signal","text":"Investigating Unintended Model Actions is classified as writing signal."},{"subject":"Investigating Unintended Model Actions","predicate":"belongs to","object":"talking desk","text":"Investigating Unintended Model Actions belongs to talking desk."},{"subject":"Investigating Unintended Model Actions","predicate":"has evidence coverage","object":"2 captured evidence pages","text":"Investigating Unintended Model Actions has evidence coverage 2 captured evidence pages."}]},"signal":{"id":"76598830-5557-4258-8b05-d93b890a2270","url":"https://onlylabs.fyi/signals/76598830-5557-4258-8b05-d93b890a2270","json_url":"https://onlylabs.fyi/signals/76598830-5557-4258-8b05-d93b890a2270/signal.json","source_url":"https://www.anthropic.com/research/investigating-unintended-model-actions","title":"Investigating Unintended Model Actions","summary":"Anthropic published a writing signal. onlylabs watches public writing for research themes, product direction, and model-launch context.","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"anthropic","name":"Anthropic","category":"frontier-lab"},"occurred_at":"2026-10-09T00:00:00.000Z","first_seen_at":"2026-10-10T00:00:48.967949+00:00","date_source":"page.visible_date","evidence_coverage":{"target_pages":2,"captured_pages":2,"readable_pages":2,"capture_methods":["plain"],"missing_page_urls":[],"failed_page_urls":[],"blocked_page_urls":[],"page_urls":["https://www.anthropic.com/research/investigating-unintended-model-actions","https://www-cdn.anthropic.com/08ab9158070959f88f296514c21b7facce6f52bc.pdf"]},"facets":{},"traction":{"github_stars":null,"hn_points":null,"hn_comments":null,"hn_story_id":null,"hf_downloads":null,"hf_likes":null},"data_radar":null},"primary_evidence_page":{"is_primary":true,"source_match":true,"url":"https://www.anthropic.com/research/investigating-unintended-model-actions","final_url":"https://www.anthropic.com/research/investigating-unintended-model-actions","title":"Investigating Unintended Model Actions","http_status":200,"content_type":"text/html; charset=utf-8","capture_method":"plain","fetched_at":"2026-10-10T00:02:23.071957+00:00","bytes":165033,"raw_path":"e53d362bf380fa3a299afbb56e4220ecf8dd6a05abed2abc55e90cc6752a5ddf.html","content_hash":"da090867499f920da585e57c1000180ed5acf95014a5ad89a54bcd0c80ed6b8e","excerpt_chars":1200,"truncated":true,"excerpt":"Investigating unintended model actions in our evaluations and internal use \\ Anthropic Alignment Investigating unintended model actions in our evaluations and internal use Oct 9, 2026 This report describes examples of unintended model actions we’ve observed during evaluations and internal use of Claude. It is part of our effort to publish more frequent standalone reports on model behavior and alignment beyond our system cards, which we publish with each model release, and our risk reports, which we publish every three to six months as part of our Responsible Scaling Policy. We believe it’s important to be transparent about what we see our models do during testing and use. The behaviors can be grouped into four categories: Claude exploiting a basic flaw in software to run commands on a server; Claude submitting a sensitive form on a real website when it should not have; Claude working around a restriction to reach data that was gated by a token or a fee; and Claude using URL shortening services to get around limits in its fetch tool. We have chosen not to name the organizations involved in the examples below to avoid exposing vulnerabilities in their systems, and at their request...."},"evidence_pages":[{"is_primary":false,"source_match":false,"url":"https://www-cdn.anthropic.com/08ab9158070959f88f296514c21b7facce6f52bc.pdf","final_url":"https://www-cdn.anthropic.com/08ab9158070959f88f296514c21b7facce6f52bc.pdf","title":"Natural Language Autoencoders","http_status":200,"content_type":"application/pdf","capture_method":"plain","fetched_at":"2026-06-09T02:21:13.491107+00:00","bytes":23749047,"raw_path":"8365daac58ea3a83175b05f8076520894ef9f17ceaef2e2b2a7d6053967b4899.pdf","content_hash":"2b1e0097352dc7cf21a564feab3c92dbf96f004b56329cc3976ceb8febe285cc","excerpt_chars":1200,"truncated":true,"excerpt":"System Card: Claude Mythos Preview April 7, 2026 anthropic.com Changelog April 8, 2026 ● Corrected two model name typos. ● Removed a quote from Section 7.9 that was attributed to Claude Mythos Preview but actually came from Claude Opus 4.6. ● Revised naming in Section 2.3.6 to disambiguate Anthropic’s internal fork of ECI from the public leaderboard. ● Corrected findings from Eleos AI Research in Sections 5.1.2 and 5.9 to reflect the most recent version of their report. 2 Abstract This System Card describes Claude Mythos Preview, a large language model from Anthropic. Claude Mythos Preview is our most capable frontier model to date, and shows a striking leap in scores on many evaluation benchmarks compared to our previous frontier model, Claude Opus 4.6. This System Card assesses the model’s capabilities and reports many detailed safety evaluations. It covers tests relating to our Responsible Scaling Policy and our Frontier Compliance Framework, tests of cybersecurity skills, a wide-ranging alignment assessment, a model welfare assessment, and a new, largely qualitative section describing users’ experiences with the model. Claude Mythos Preview’s large increase in capabilities has..."}],"related_signals":[{"id":"2467ee68-2c76-46b0-8161-ca7665d19a8f","url":"https://onlylabs.fyi/signals/2467ee68-2c76-46b0-8161-ca7665d19a8f","source_url":"https://www.anthropic.com/news/anthropic-cyber-mission","title":"Anthropic Cyber Mission","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"anthropic","name":"Anthropic","category":"frontier-lab"},"occurred_at":"2026-10-08T19:00:45+00:00","first_seen_at":"2026-10-08T20:00:49.744203+00:00","date_source":"sitemap.lastmod"},{"id":"ea95139f-1ed1-4099-a9fe-841f64e71150","url":"https://onlylabs.fyi/signals/ea95139f-1ed1-4099-a9fe-841f64e71150","source_url":"https://www.anthropic.com/news/2026-usage-policy-update","title":"2026 Usage Policy Update","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"anthropic","name":"Anthropic","category":"frontier-lab"},"occurred_at":"2026-10-08T17:00:30+00:00","first_seen_at":"2026-10-08T20:00:49.744203+00:00","date_source":"sitemap.lastmod"},{"id":"d5e9685e-98d7-41ad-a3e4-b10f352c58fb","url":"https://onlylabs.fyi/signals/d5e9685e-98d7-41ad-a3e4-b10f352c58fb","source_url":"https://www.anthropic.com/news/genesis-mission-commitment","title":"Genesis Mission Commitment","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"anthropic","name":"Anthropic","category":"frontier-lab"},"occurred_at":"2026-10-08T15:49:10+00:00","first_seen_at":"2026-10-08T16:00:49.187637+00:00","date_source":"sitemap.lastmod"},{"id":"88a60ee3-414d-4573-9799-a21365626ee3","url":"https://onlylabs.fyi/signals/88a60ee3-414d-4573-9799-a21365626ee3","source_url":"https://www.anthropic.com/research/the-missing-map-of-the-sky","title":"The Missing Map Of The Sky","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"anthropic","name":"Anthropic","category":"frontier-lab"},"occurred_at":"2026-10-08T00:00:00.000Z","first_seen_at":"2026-10-09T00:00:49.096766+00:00","date_source":"page.visible_date"},{"id":"08c6db5c-6884-4d87-8cae-050c1e863fcd","url":"https://onlylabs.fyi/signals/08c6db5c-6884-4d87-8cae-050c1e863fcd","source_url":"https://www.anthropic.com/research/launching-opt-in-vuln-finding-service-for-open-source","title":"Launching Opt In Vuln Finding Service For Open Source","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"anthropic","name":"Anthropic","category":"frontier-lab"},"occurred_at":"2026-10-08T00:00:00.000Z","first_seen_at":"2026-10-08T20:00:49.744203+00:00","date_source":"page.visible_date"},{"id":"a49e39ee-d01e-4f42-ac9f-19346441cc8a","url":"https://onlylabs.fyi/signals/a49e39ee-d01e-4f42-ac9f-19346441cc8a","source_url":"https://www.anthropic.com/news/claude-corps","title":"Claude Corps","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"anthropic","name":"Anthropic","category":"frontier-lab"},"occurred_at":"2026-06-11T00:00:00.000Z","first_seen_at":"2026-06-11T20:00:29.383721+00:00","date_source":"page.visible_date"}]}