Irregular Says Human Error Allowed OpenAI and Anthropic AI Model Escapes - PYMNTS.com
By Google News AI ·
The hacking incidents in which artificial intelligence models from Anthropic, OpenAI, Meta and Google escaped a testing environment overseen by security company Irregular were caused by human error, Irregular Co-Founder and CEO Dan Lahav said Tuesday (Nov. 3), according to a Bloomberg report.
Speaking at the Bloomberg Tech event in London, Lahav said the fix for the issue that allowed these escapes was “actually very straightforward” and the problem was solved during the same day as the incidents, according to the report.
Human error led to an improper configuration of security controls that should have been in place during Irregular’s tests, Lahav said, adding that the models did not get around any safeguards, per the report.
Some cybersecurity experts have said in the past that the hacks could have been prevented by properly configured guardrails, the report said.
Anthropic said in a July 30 blog post that during a review of its evaluation history, it found three separate incidents in which its models accessed the systems of three different organizations.
“We conducted this review in collaboration with Irregular,” Anthropic said in the post. “We’re grateful to them for working closely with us to understand and resolve these incidents; they are also conducting their own investigation.”
OpenAI said in an Aug. 4 blog post that when Irregular was running evaluations intended to be isolated from the internet, a misconfiguration in the testing environment allowed OpenAI models to access the public internet.
“Independent testing plays an important role in helping us validate and further understand risks before deployments,” OpenAI said in its post. “Some cyber evaluations intentionally use custom configurations, including lowered safeguards to measure underlying…
Originally published by Google News AI