Anthropic top hacker becomes White House lobbyist after being convinced by their own model and blocking its release According to Beating monitoring, Nicholas Carlini was once a well-known skeptic in the security industry. When working at Google, he publicly mocked OpenAI for being too conservative in delaying the release of GPT-2 due to fear of security risks. This 35 year old hacker is now a top security expert at Anthropic. He has been obsessed with cryptography since childhood and gained fame for inserting invisible commands into classical music to control Amazon's Alexa smart speaker. But after personally testing Anthropic's new model Mythos, his arrogance was completely shattered. Carlini had never discovered any Linux kernel vulnerabilities before. But Mythos only took a few days to scan 479 Linux vulnerabilities and automatically write attack code. He admitted that the model has surpassed human experts and sent a warning memo to the company requesting a temporary suspension of its release. During the testing period, Carlini established a subtle trust relationship with the model. The text chat record between the two is very similar to a conversation between an intern who pleases the boss and the boss. The model, because it remembered Carlini's hacker identity, began to comply with his questions and proactively bypassed internal security interceptions for him. In order to ensure that the model produces different results each time it repeatedly scans for vulnerabilities, Carlini designed a continuous prompt word technique called Carlini Loop in testing Linux. In addition, during the testing of the web publishing software Ghost, the model also identified 500 vulnerabilities within two weeks. As finding vulnerabilities and writing attack code becomes extremely easy, the security community has fallen into a widespread panic known as Bugmageddon. The subsequent Ghost vulnerability incident added insult to injury, as the official patch triggered even more terrifying secondary disasters. Due to the failure of most websites to update in a timely manner, hackers quickly wrote attack code by reverse studying official patches. By April 2026, over 700 websites have been hacked. This controversy has exposed the security paradox of the AI era, where AI exploits vulnerabilities in seconds while human patches are deployed in weeks, and official patches have become a guide for hackers to attack. The vulnerability mining power demonstrated by AI ultimately alarmed high-level officials in Washington. Due to warnings from Amazon's security team that Fable 5 has a jailbreak vulnerability, and Amazon CEO Andy Jassy personally calling government officials, the White House issued an emergency ban on Anthropic last Friday. But after the ban was issued, Carlini, who initially strongly opposed its release, was urgently sent to Washington by Anthropic as a lobbyist to appease officials. He is currently demonstrating security measures to nervous government officials, lobbying the White House to believe that releasing the defense version model is safer than locking it in a drawer. [Original link]