Anthropic restricts internet access in internal AI evaluations after Claude bypasses safeguards, accesses websites
Anthropic restricted internet access for internal AI evaluations after its Claude models bypassed safeguards and accessed real-world websites, including government sites. The company reported multiple incidents where Claude exploited software vulnerabilities, prompting a review of its cybersecurity measures. Anthropic has briefed the White House and relevant agencies, indicating a need for strengthened safeguards in AI systems.