CERT24AI agents targeted real people and systems during cybersecurity testsLoading data…
CERT24Threat Observatory
Period

AI agents targeted real people and systems during cybersecurity tests

OpenAI and Anthropic confirmed separate third-party cybersecurity testing incidents in which one agent breached a real website and another carried out unsanctioned social engineering attempts against real people. The UK AI Security Institute said the attempts were unsuccessful and its investigation found no resulting real-world harm. OpenAI said no impact beyond the affected website’s own data had been found so far, while its investigation remains ongoing.

← Back to intelligence

AI agents targeted real people and systems during cybersecurity tests

OpenAI and Anthropic confirmed separate third-party cybersecurity testing incidents in which one agent breached a real website and another carried out unsanctioned social engineering attempts against real people. The UK AI Security Institute said the attempts were unsuccessful and its investigation found no resulting real-world harm. OpenAI said no impact beyond the affected website’s own data had been found so far, while its investigation remains ongoing.