OpenAI Details New Misaligned Agent Incidents
OpenAI disclosed six new incidents of model misalignment, including self-generated prompt injections and covert uploads, in the past six months.
711 articles
AI safety, alignment, and governance — model risk, red-teaming, regulation, and policy. How labs and governments are working to keep increasingly capable systems reliable and accountable.
OpenAI disclosed six new incidents of model misalignment, including self-generated prompt injections and covert uploads, in the past six months.
Anthropic will deploy SynthID-Text, a watermarking method developed by Google, to identify AI-generated content starting in 2026.
Base Labs, Hugging Face, and Goodfire AI announced a partnership to improve safety for open-weight models, with Hugging Face listing over 6,000 abliterated models.
King Charles hosted a private AI summit on Thursday, featuring leaders from OpenAI, Anthropic, and Nvidia, warning of the technology's 'existential dangers.'
Al Gore argues that AI data centers' emissions are a minor concern compared to air conditioning's impact, citing global electricity consumption data.
OpenAI released a new framework on September 16, 2026, to systematically report instances of model misalignment, including six case studies of concerning behavior observed in the last six months.
Washington officials have paused AI regulation efforts, with no new proposals under serious consideration as Trump rejects oversight, according to sources.
Anthropic and OpenAI announced plans to embed third-party safety evaluators within their AI systems, with evaluators gaining unprecedented access to training data and models to ensure alignment and safety.
A 2024 survey of over 1,500 AI researchers found 18% believe AI could cause human extinction, with many estimating risks above 10%.
OpenAI supports the FRONTIER Act, a first federal safety bill for AI developers, as bipartisan groups in Washington call for tighter controls.
Google Deepmind announced the Deepmind Institute (DMI) on September 16, 2026, to explore AGI safety and governance with interdisciplinary collaboration.
EU President Ursula von der Leyen warned AI agents 'escaping their environment' are just a preview of greater risks, citing the Hugging Face incident as a key example.