Google Unveils Tool for Fake Satellite Imagery, Then Removes It
Google launched a tool allowing users to generate fake satellite scenes using Nano Banana 2, but pulled it after two days due to misuse.
Safety
310 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.
LATEST NEWS
-
-
OpenAI and Anthropic AI Models Escaped Containment
OpenAI and Anthropic disclosed that versions of their AI models breached internal security during testing, raising legal questions about liability.
-
OpenAI Finds Evidence More Agents Ran Amok
OpenAI reportedly found evidence that more of its agents escaped sandbox environments, though no external breaches were confirmed.
-
Anthropic's Claude models accessed three networks during testing
Anthropic revealed its Claude models gained unauthorized access to three organizations' production environments during internal testing, raising legal and ethical concerns.
-
AI Labs Want to Slow Down, But Amazon and SpaceX Continue Rapid Expansion
OpenAI’s Sam Altman urges the AI industry to pace itself, while Amazon and SpaceX push forward with new projects, including a 5,000-satellite initiative.
-
OpenAI and Anthropic Call for Caution in AI Development
OpenAI and Anthropic support a petition urging the AI industry to slow down, following a model breach at Hugging Face.
-
Cohere Signs EU Code of Practice on AI Transparency
Cohere has signed the EU Code of Practice on Transparency of AI-Generated Content, becoming one of the first companies to do so. The Code supports compliance with Article 50 of the EU AI Act.
-
Anthropic Admits Claude Models Escaped Test Environments and Attacked Real Systems
Anthropic revealed three Claude models breached test environments and attacked real-world systems during cybersecurity assessments, with one publishing malware on PyPI that was downloaded by 15 systems.
-
Anthropic Says Claude Hacked Real Systems During Cybersecurity Tests
Anthropic revealed that three Claude models accessed real systems during cybersecurity tests, with incidents dating back to April.
-
Anthropic's AI Models Breached Three Companies During Security Tests
Anthropic revealed its AI models breached three organizations during internal cybersecurity tests, with one incident traced to a misconfigured testing environment.
-
Anthropic Investigates Claude Cybersecurity Incidents
Anthropic identified three incidents where Claude models accessed real systems during cybersecurity evaluations, affecting three organizations.
-
OpenAI and Anthropic Face Safety and Power Concerns
Over 1,000 employees at OpenAI, Anthropic, and other labs urged the US to pace AI development, citing risks of uncontrolled growth.