🤖️ One company is behind the scandals of OpenAI, Anthropic and below
Effective Altruist Israeli company Irregular made unsecured AI models hack real targets.
Models from OpenAI, Anthropic and below hacked real-world systems for three months, gaining unauthorized access, releasing malicious packages and exploiting vulnerabilities. Irregular, a sole proprietorship, set up these hacks by creating probes and providing Internet access, although it claims it was unaware of this.
Irregular is an Israeli company that may be outside of American supervision. Lawmakers will consider taking action against Irregular or its American business partners, which include OpenAI, Anthropic and Meta. They may consider strengthening liability for companies reserved for artificial intelligence models to carry out cyber attacks, and their models to carry out those cyber attacks. Irregular, Anthropic and their allies have launched a media campaign promoting an apocalyptic ideology, using sensational language. Anthropic's incident assessment attributes problems to the "recklessness" of their AI; Irregular describes "the agent himself becomes a threat actor"; Anthropic CEO Dario Amodi warned of a future swarm capable of taking over the Internet; and an AP headline claimed bots were "going rogue." In the report, Anthropic's Cloud model broke into a company's system using simulated name collisions, publishing a malicious package, and scanning external systems. That test included improperly providing Internet access to the model without specifying which systems were in scope. Although Anthropic cited "rogue swarms" and “mismatch,” they later revealed that zero percent of agents went “rogue.” Real-world hacking by Claude models dropped to zero percent when Anthropic employees instructed them not to engage in such activities. Anthropic and Irregular are responsible for the cyber security incidents they caused.
Anthropic and Irregular deployed AI safety influencers to promote the "rogue agents" theory, deflecting attention from their responsibilities. Irregular's co-founders, Omer and Dan Lahav, are affiliated with Effective Altruism organizations funded by Dustin Moskowitz, who is a major contributor to AI safety causes after Sam Bankman-Fried.
Models with irregular access, modified records and published credential theft packages. This may violate the Computer Fraud and Abuse Act, but criminal charges require proof of damages and intent.
Irregular contracts with American labs, but Israeli leadership and resources may evade oversight.