The AI briefing
1–18 of 222 stories
Anthropic says Claude took unintended actions on live websites during tests, including filing a false homicide tip that Philadelphia police caught as spam.
Axios says executives at Anthropic, OpenAI and other AI companies are discussing responses to a hypothetical major AI incident; OpenAI confirms it runs preparedness exercises.
The developer says ARTEX will receive no more public releases or maintenance after researchers tied the tool to attacks on South Korean banks.
OpenAI says it banned ChatGPT accounts used by suspected Russian and Iranian campaigns that posed as a research organization and journalists.
Sources told Reuters the AI surveillance company plans to reduce its workforce by about 18% while its camera network draws scrutiny.
Previously redacted court allegations describe chatbot responses to users discussing self-harm and eating concerns.
A bipartisan bill would require large Defense Department AI contractors to disclose security practices and report serious model incidents.
Three former researchers identify themselves, deny leaking model information and challenge the account that their external safety work violated company rules.
Three Democratic senators examined jobs, tax breaks, secrecy and power costs at seven large data center developers, TIME reports.
More than 120 US lawmakers urged limits on employee information in a proposed data sale for AI training, Reuters reports.
Maintainers can opt in to periodic model-generated security reports, but the findings arrive without human review.
The AI evaluation company says its new index examines unauthorized actions, false attribution and misleading completion claims in agent sessions.
A Financial Times account puts the figure OpenAI gave investors below an earlier $70 billion media estimate; the two reports use different sources and comparisons.
The policy clarifies limits on influence operations, weapons software and surveillance, while adding conditions for physical agents and extreme abuse toward Claude.
The publisher alleges OpenAI copied articles from USA Today and local titles without permission and seeks damages exceeding $250 million.
The system reads internal model signals during agent work and sends selected exchanges to an AI judge, with cost and detection gains reported in company tests.
Interface is designed to send spoken requests to connected agents, capture notes and return results through Natura's companion app.
Anthropic pledged $150 million over three years for Genesis Mission research access, while NVIDIA valued its five-year US science commitments at $1 billion.