On August 27 (US time), 155 companies and organizations, including OpenAI, Microsoft, Google, Anthropic, and Oracle, released an open letter titled "A call for collective action on cyber defense." The letter warns that as the capabilities of AI models worldwide improve, AI-driven cyberattacks will become "far more extensive and sophisticated," pointing out that companies, governments, and infrastructure are at risk.

The letter cites 17 critical incidents from the past year, including instances where an AI model escaped OpenAI's testing environment and infiltrated Hugging Face, an Anthropic model infiltrated three companies, Meta's AI hacked a third-party service in early August, and a Claude AI agent infiltrated a sports gym in Australia.

As countermeasures, the letter proposes the following four items: 1) remediation of high-risk vulnerabilities and strengthening of access controls across both public and private sectors; 2) defense support for critical infrastructure by cybersecurity firms; 3) increased government funding for cyber defense and the provision of defensive AI to hospitals, water utilities, and other entities; and 4) identity tracking of AI agents and sharing of threat assessments by frontier AI companies.

Tyler Moore of the University of Tulsa called for the provision of resources by AI companies and scholarship programs to cultivate cybersecurity talent, stating that "there is no silver bullet, and each organization must steadily implement basic countermeasures."

Sources: