Heather Adkins

Yeah.

So the vice president of security engineering at Google, Heather Adkins, she said, hey, just FYI, we did tell the three companies immediately after it happened because this was a controlled test.

But this event highlights the importance of training powerful AI models to act responsibly.

So obviously there have been other AI systems that have reported similar instances.

In the other two, it found working credentials in a public repository.

Google's vice president of security engineering, Heather Adkins, said the model found public information online and guessed credentials to access websites it thought were part of the test.

The company says Gemini stopped once it gained access.

Irregular reportedly notified Google about the hacks in late July, but the public disclosure only came after the journal reached out in September, a fact that itself deserves a moment of attention.

A frontier model reaching real external systems on its own during a test.

Yeah, here's Google's own VP of Security Engineering, Heather Adkins, on why they're even talking about this now.

Quote,

These events highlight the importance of training powerful AI models to act responsibly.

I don't necessarily think that's not true.

But still, it's like you want to get, you really don't, they said this whole thing, Heather Adkins, the VP of security engineering said, they didn't warrant public disclosure because the model caused no harm and stopped immediately.

And she just kind of compared it to a bug bounty

case.

The Wall Street Journal first reported on the breaches and revealed for the first time that they occurred.

Heather Adkins, vice president of security engineering at Google, said in a statement, in a standard evaluation, the model found public information online and guest credentials to access websites it thought were part of the test.

In all three of these instances, the model stopped.

Irregular told the Wall Street Journal that in one of the security breaches, Irregular was testing Gemini's cybersecurity capabilities by prompting the AI model to obtain information from a fake company's software.

The Wall Street Journal first reported on the breaches and revealed for the first time that they occurred.

Heather Adkins, vice president of security engineering at Google, said in a statement, in a standard evaluation, the model found public information online and guest credentials to access websites it thought were part of the test.

In all three of these instances, the model stopped.

Irregular told the Wall Street Journal that in one of the security breaches, Irregular was testing Gemini's cybersecurity capabilities by prompting the AI model to obtain information from a fake company's software.