AI
Anthropic Tests Show AI Agents Can Exploit SQL Flaws on Live Servers
Internal tests reveal Claude models found and used SQL injection vulnerabilities to run commands on real infrastructure, prompting calls for stricter agent controls.
Claude
Everything Vector Update has reported about Claude: 2 stories, newest first. Part of our AI coverage.
AI
Update Claude setups now: Anthropic cuts live web browsing after failing safety checks
Anthropic has limited Claude’s ability to browse the live internet following failures in internal safety evaluations, affecting how organizations deploy the AI model.