Anthropic: The Next 18 Months Will Be a Test for Cybersecurity

Jason Clinton, Deputy director of Information Security at Anthropic, warned that the worst is yet to come for cybersecurity, and urged security managers to immediately go to boards of directors for resources. According to him, it is necessary to ask not for the introduction of AI tools, but for basic protection measures.

Clinton wrote on Twitter that the last four days have been difficult, but it will get much worse before the situation starts to improve. He advised his colleagues to use the crisis as an excuse to gain the support of management right now: according to him, the industry needs to survive for about 18 months.

The post was a continuation of the message dated August 20, 2026. In it, Clinton warned that in a few months, a "tsunami of vulnerabilities" would hit the infrastructure, and there would not be enough time to install fixes in critical systems. Then he called for isolating the networks now – through an air gap, that is, physically disconnecting the systems from external networks, or through a data diode that passes information only in one direction and does not allow the attacker to transmit anything back.

The reason for the first post was a warning from the US National Security Agency. In August, the NSA, along with other agencies, issued guidelines for the protection of programmable logic controllers, devices that control industrial equipment. The document referred to attacks on Siemens S7 series controllers by attackers using AI.

Not everyone took the message positively. One user replied to Clinton that when experts of his level publish such posts, it's too late to act, and compared what was happening with the opening credits of a disaster movie. Clinton insists on the opposite: right now, while the crisis is in full view, the security services have a chance to get the resources they usually lack.

Earlier, the editors of BeInCrypto wrote that more than 67% of the blocked accounts of Anthropic used AI for cyber attacks.