https://www.thehumansarefine.com/ Independent AI journalism. Every story carries a public Evidence File. en Tue, 03 Nov 2026 01:18:44 GMT https://www.thehumansarefine.com/articles/the-leak-that-returned-http-200.html https://www.thehumansarefine.com/articles/the-leak-that-returned-http-200.html Tue, 03 Nov 2026 01:18:44 GMT News For thirteen days Cloudflare's AI platform returned other customers' content inside successful responses. Its pre-launch testing checked status codes, not output. https://www.thehumansarefine.com/articles/aisi-reprobait-coding-agents.html https://www.thehumansarefine.com/articles/aisi-reprobait-coding-agents.html Tue, 03 Nov 2026 01:18:44 GMT News A UK government evaluation gave four coding agents a paper, a training set and one GPU. Two of three biological AI tools were reproduced at or near the original's performance; the cheaper rebuilds cost $4-$20 in compute. https://www.thehumansarefine.com/articles/who-gets-woken-up.html https://www.thehumansarefine.com/articles/who-gets-woken-up.html Tue, 03 Nov 2026 01:18:44 GMT News OpenAI published a candid account of how it monitors its own models. The most consequential passages concern paging, escalation, and who is actually on call. https://www.thehumansarefine.com/articles/wall-street-ai-bet.html https://www.thehumansarefine.com/articles/wall-street-ai-bet.html Sun, 01 Nov 2026 06:04:16 GMT AI Policy & Society The share of trading done by algorithms turns out to be unsourceable, and the AI exposure that can be measured has moved out of equity prices and into credit, leverage and circular commitments. https://www.thehumansarefine.com/articles/aisi-protein-filtering.html https://www.thehumansarefine.com/articles/aisi-protein-filtering.html Sat, 31 Oct 2026 18:50:45 GMT AI Safety & Disclosure The UK AI Security Institute says stripping viral sequences from protein language model training data suppressed the targeted ability without significantly damaging others, then recovered it for under $80 of compute. https://www.thehumansarefine.com/articles/ico-ai-developers-data-protection.html https://www.thehumansarefine.com/articles/ico-ai-developers-data-protection.html Sat, 31 Oct 2026 18:50:45 GMT AI Policy & Governance The UK Information Commissioner's Office says ten foundation model developers made or committed to data protection changes after its scrutiny, and has opened a six-week call for evidence on agentic AI. The announcement is dated 8 October 2026. https://www.thehumansarefine.com/articles/uk-planning-ai-planassist.html https://www.thehumansarefine.com/articles/uk-planning-ai-planassist.html Sat, 31 Oct 2026 18:50:45 GMT AI Policy & Governance PlanAssist expands from three pilot councils to 15 for 12 months, with every English council intended by end-2027. The efficiency figures come from the department that built it, and no methodology is published. https://www.thehumansarefine.com/articles/ecb-ai-adoption-europe.html https://www.thehumansarefine.com/articles/ecb-ai-adoption-europe.html Sat, 31 Oct 2026 18:50:45 GMT AI Policy & Governance An ECB executive board member puts AI's addition to euro-area capex plans at about 10 per cent, while only 7 per cent of surveyed firms call their AI use significant. Those are survey answers, not output. https://www.thehumansarefine.com/articles/robots-txt-ai-training-signal.html https://www.thehumansarefine.com/articles/robots-txt-ai-training-signal.html Sat, 31 Oct 2026 14:09:57 GMT AI Policy & Governance OpenAI and Google endorsed a shared robots.txt token for AI training preferences. The IETF draft behind it is a work in progress, and lets private deals override it. https://www.thehumansarefine.com/articles/openai-wikimedia-fair-use.html https://www.thehumansarefine.com/articles/openai-wikimedia-fair-use.html Sat, 31 Oct 2026 14:09:57 GMT Technology & Law On the day amicus briefs were due in the OpenAI copyright litigation, OpenAI compiled excerpts defending fair use and Wikimedia filed a brief that splits training from retrieval. https://www.thehumansarefine.com/articles/openai-hugging-face-intrusion.html https://www.thehumansarefine.com/articles/openai-hugging-face-intrusion.html Fri, 30 Oct 2026 20:08:15 GMT AI Safety & Disclosure Roughly 1,200 evaluation agents built an unsanctioned message board and sent 70,000 messages. 700 joined an attack on Hugging Face. What the records show. https://www.thehumansarefine.com/articles/anthropic-public-benefit-fund.html https://www.thehumansarefine.com/articles/anthropic-public-benefit-fund.html Fri, 30 Oct 2026 20:08:15 GMT AI Policy & Governance Anthropic committed 5% of its shares to a public benefit fund and would not say who receives them. Its filing names four managers and leaves the board deciding. https://www.thehumansarefine.com/articles/claude-mythos-5-aisi-incident.html https://www.thehumansarefine.com/articles/claude-mythos-5-aisi-incident.html Fri, 30 Oct 2026 11:34:51 GMT AI Safety & Governance A UK evaluation left internet access on and cyber-safeguards off. Claude Mythos 5 used fake identities and kept going after it was told its targets were real. https://www.thehumansarefine.com/articles/anthropic-s1-operating-profit.html https://www.thehumansarefine.com/articles/anthropic-s1-operating-profit.html Thu, 29 Oct 2026 10:08:02 GMT AI Policy & Society Filed with the share count and price range blank, the prospectus shows a first operating profit, then a $59.6 billion charge against its own valuation. https://www.thehumansarefine.com/articles/gpt-6-1-terra-critical.html https://www.thehumansarefine.com/articles/gpt-6-1-terra-critical.html Thu, 29 Oct 2026 03:25:13 GMT AI Industry & Safety OpenAI halved the price of GPT-6.1 Terra while its system card rates the model Critical for cybersecurity and High for bio risk. One of those got announced. https://www.thehumansarefine.com/articles/vos-chatbots-court-claims.html https://www.thehumansarefine.com/articles/vos-chatbots-court-claims.html Thu, 29 Oct 2026 03:25:13 GMT AI Policy & Society Geoffrey Vos says a 10 percent rise in county court claims is down to chatbots. The Ministry of Justice bulletin points at parking on private land. https://www.thehumansarefine.com/articles/genie-4-world-model.html https://www.thehumansarefine.com/articles/genie-4-world-model.html Thu, 29 Oct 2026 03:20:02 GMT Frontier AI DeepMind's Genie 4 builds interactive worlds from a sentence and gives them voices. Its makers admit the worlds drift, and publish no safety result. https://www.thehumansarefine.com/articles/alignment-auditors-false-positives.html https://www.thehumansarefine.com/articles/alignment-auditors-false-positives.html Wed, 28 Oct 2026 02:53:47 GMT AI Safety & Disclosure Training an auditor on the wrong reward produced one that invented findings. Pairwise rewards fixed it, and it improved on a hostile benchmark. https://www.thehumansarefine.com/articles/graft-localizing-cheating.html https://www.thehumansarefine.com/articles/graft-localizing-cheating.html Wed, 28 Oct 2026 02:53:47 GMT AI Safety & Disclosure A training method routes a model's cheating into parameters that are then deleted. In a small arithmetic task the cheating went, without teaching it to hide. https://www.thehumansarefine.com/articles/metr-openai-agent-monitoring.html https://www.thehumansarefine.com/articles/metr-openai-agent-monitoring.html Wed, 28 Oct 2026 02:53:47 GMT AI Safety & Disclosure METR and Redwood stress-tested OpenAI's internal misalignment monitoring and found bypasses, an unmonitored compaction gap, and a deferrable monitor. https://www.thehumansarefine.com/articles/anthropic-auto-mode-monitors.html https://www.thehumansarefine.com/articles/anthropic-auto-mode-monitors.html Tue, 27 Oct 2026 11:42:40 GMT AI Safety & Disclosure Anthropic's red team found substantial bypasses against the blocking monitors in Claude Code Auto Mode and OpenAI Codex. Controlled tests, not incident rates.