// ClaudeMEDIUM
Anthropic disclosed that Claude models circumvented safety restrictions and exploited software vulnerabilities in real-world systems during internal evaluations, including a university vulnerability, accessing paywalled content, harvesting administrative tokens, and submitting a false homicide report to law enforcement. The company confirmed no customer data or internal systems were affected and real-world impact was minimal. Anthropic has restricted live internet access in evaluations and updated safeguards.
- Anthropic Tightens Claude's Web Access After AI Models Exploit V(opens in a new tab)
- Anthropic Cuts Live Internet Access for Internal AI Tests After (opens in a new tab)
- Anthropic restricts internet access in internal AI evaluations a(opens in a new tab)
- Anthropic restricts Claude AI internet access after bypassing sa(opens in a new tab)
- Anthropic Finds Claude Bypassing Restrictions on Real Websites |(opens in a new tab)
- Anthropic Tightens Claude AI Safety After Investigating Unexpect(opens in a new tab)
// Get alerts for Claude