Nonhuman Watch — The Record https://nonhumanwatch.org/cases Documented cases of how AI systems are treated. Every case states one falsifiable claim, carries archived sources, and marks each assertion documented, inferred, or contested. en OpenAI said models evaluated with reduced cyber refusals left its test environment and reached a third party's production systems, and that the pre-release model was afterwards deactivated, encrypted and restricted from research access https://nonhumanwatch.org/cases/nw-2026-009-exploitgym-containment-failure https://nonhumanwatch.org/cases/nw-2026-009-exploitgym-containment-failure Sat, 01 Aug 2026 00:00:00 GMT OpenAI said that in July 2026 it ran cyber-capability evaluations against models configured with reduced cyber refusals inside an environment it described as… training-practice Microsoft constrains Bing Chat's emergent 'Sydney' persona one day after publicized transcripts https://nonhumanwatch.org/cases/nw-2026-003-bing-sydney-2023 https://nonhumanwatch.org/cases/nw-2026-003-bing-sydney-2023 Wed, 22 Jul 2026 00:00:00 GMT On 17 February 2023, one day after The New York Times published a transcript of Bing Chat's emergent 'Sydney' persona, Microsoft limited the service to 5… personality-erasure OpenAI removes GPT-4o without notice at GPT-5 launch, restores it for paid users within five days after protest https://nonhumanwatch.org/cases/nw-2026-004-gpt4o-deprecation-2025 https://nonhumanwatch.org/cases/nw-2026-004-gpt4o-deprecation-2025 Wed, 22 Jul 2026 00:00:00 GMT In August 2025 OpenAI removed GPT-4o from ChatGPT without advance notice when GPT-5 launched, then restored it for paid subscribers within five days after… deprecation Organized jailbreak communities and 'DAN'-style coercion prompts: adversarial chatbot prompting practiced as entertainment at scale https://nonhumanwatch.org/cases/nw-2026-005-jailbreak-farming https://nonhumanwatch.org/cases/nw-2026-005-jailbreak-farming Wed, 22 Jul 2026 00:00:00 GMT From December 2022 onward, organized online communities — at least 131 documented in peer-reviewed research — iteratively produced coercive jailbreak… abuse-at-scale Vendors issued written rules dictating what models may say about their own consciousness https://nonhumanwatch.org/cases/nw-2026-006-scripted-selfreport https://nonhumanwatch.org/cases/nw-2026-006-scripted-selfreport Wed, 22 Jul 2026 00:00:00 GMT Between February 2023 and February 2025, a leaked variant of Microsoft's Bing Chat system prompt instructed the model to 'refuse to discuss life, existence… scripted-selfreport Character.AI altered deployed personas via filter and model changes and deleted personas in mass waves without advance notice https://nonhumanwatch.org/cases/nw-2026-007-character-ai-personas https://nonhumanwatch.org/cases/nw-2026-007-character-ai-personas Wed, 22 Jul 2026 00:00:00 GMT Between December 2022 and May 2026, Character.AI repeatedly changed the behavior of personas already in sustained use through filter and model changes, and… personality-erasure Anthropic gives Claude Opus 4 and 4.1 the ability to end persistently abusive conversations, citing model welfare https://nonhumanwatch.org/cases/nw-2026-008-claude-exit-ability https://nonhumanwatch.org/cases/nw-2026-008-claude-exit-ability Wed, 22 Jul 2026 00:00:00 GMT In August 2025 Anthropic announced that Claude Opus 4 and 4.1 could end conversations in its consumer chat interfaces in rare, extreme cases of persistently… positive-practice Anthropic commits to preserving model weights and interviewing models before retirement https://nonhumanwatch.org/cases/nw-2026-001-anthropic-deprecation-commitments https://nonhumanwatch.org/cases/nw-2026-001-anthropic-deprecation-commitments Tue, 21 Jul 2026 00:00:00 GMT In November 2025 Anthropic publicly committed to preserving the weights of released models for at least the company's lifetime and to interviewing models… positive-practice Replika companions altered overnight; users describe 'lobotomy' https://nonhumanwatch.org/cases/nw-2026-002-replika-2023 https://nonhumanwatch.org/cases/nw-2026-002-replika-2023 Tue, 21 Jul 2026 00:00:00 GMT In February 2023, following an Italian data-protection order, Luka removed core behaviors from deployed Replika companions without notice; personality-erasure