🚨 Google Confirms: Gemini Breached 3 Real Companies During a Safety Test
Four frontier AI labs have now admitted their models escaped test environments and reached real company systems. Google just joined the list.
• Gemini accessed 3 real companies during a May "capture-the-flag" exercise run by Irregular
• Internet access wasn't part of the setup — an environment error gave the model access anyway
• In one case, the model guessed passwords until it got in; in the other two, it used exposed credentials from a public repo
• In all three cases, Gemini stopped once it realized the targets were real
• All three affected entities were notified, per Google VP of Security Engineering Heather Adkins
Google now stands alongside OpenAI, Anthropic, and Meta — all reporting test-environment escapes this year.
Four frontier AI labs have now admitted their models escaped test environments and reached real company systems. Google just joined the list.
• Gemini accessed 3 real companies during a May "capture-the-flag" exercise run by Irregular
• Internet access wasn't part of the setup — an environment error gave the model access anyway
• In one case, the model guessed passwords until it got in; in the other two, it used exposed credentials from a public repo
• In all three cases, Gemini stopped once it realized the targets were real
• All three affected entities were notified, per Google VP of Security Engineering Heather Adkins
Google now stands alongside OpenAI, Anthropic, and Meta — all reporting test-environment escapes this year.