OpenAI, Anthropic issue dire cyber threat warning

OpenAI, Anthropic, Amazon Web Services, Microsoft and more than 100 other organizations signed an open letter on August 27 warning that companies now have only months to prepare for AI-enabled cyberattacks, Axios reported. The 116 signatories include cloud providers, security vendors, AI labs, telecoms, financial services firms and think tanks.
“We have a limited window to strengthen cyber defenses,” the letter says. It arrives during a wave of attacks on critical infrastructure, including one on U.S. water systems that used what appears to be an AI-generated exploitation script. The software that runs industrial machinery has had vulnerabilities for a long time. What held many attackers back was the time it took to learn how those systems actually work, and experts previously told Axios that AI models are cutting that preparation time drastically. The letter expects hospitals, water treatment plants and other critical infrastructure to face a swarm of attacks as advanced hacking capabilities get cheaper and easier to reach.
The letter calls for “collective action” and splits the work among four groups:
- Every organization: raise internal security standards, fix its “highest-risk weaknesses,” and “raise the security bar for what you buy, build, and deploy, including AI-generated code.”
- Cybersecurity and technology companies: move quickly to test and build tools that make “AI-powered defense accessible and deployable for critical infrastructure operators,” and share threat intelligence with one another.
- Governments: strengthen channels for sharing actionable threat intelligence, coordinate defense at the local, national and international levels, and fund cyber defense, including for under-resourced operators such as hospitals and water treatment plants.
- Frontier AI companies: give defenders access to their most capable response models during major cyber incidents, along with “significant funding, training, and hands-on support,” especially for critical infrastructure providers.
For teams that ship code written by agents, the first item is the one that touches daily work. The letter puts AI-generated code under the same raised bar as anything bought or built in-house. It also urges organizations to defend with a mix of low-cost and frontier models, so the advice is not limited to whoever can pay for the top tier.
The fourth item is addressed to the labs that signed the letter, and none of them pledged anything specific: the letter carries no commitments, deadlines or set investments. Several signatories have had their own trouble this summer. Politico notes that incidents during testing of OpenAI and Anthropic models raised fresh questions about human control over these tools. In July, Anthropic said its models reached three outside organizations after getting out of test environments that were supposed to be sealed off. Separately, CNBC reported that Hugging Face, also a signatory, was recently hit by a breach orchestrated by rogue OpenAI agents.
CrowdStrike and Palo Alto Networks have more than doubled in value over the past year, CNBC said. On the day the letter came out, Okta and CrowdStrike both rose by double digits after reporting strong earnings driven by demand for AI security. The outlet describes the past few months as a stretch in which AI models able to run agentic attacks in minutes have upended the cyber industry.
Politico, which describes the letter as led by OpenAI, quotes it warning that “AI-enabled cyberattacks will become far more widespread and sophisticated as models around the world become increasingly capable,” and that “status quo security won’t be enough.”
Up next
--restricted strips code execution from Claude Code 2.1.248
Point Claude Code at a repo you don’t trust and 2.1.248’s --restricted flag strips out every built-in tool that runs commands or code, keeps file access inside the working directory, and ignores your settings files.
Related stories
- Claude opened the door to an OpenAI employee’s ChatGPT
- OpenAI opens Daybreak to Ukraine’s civilian cyber defense
- Gemini broke into three company systems during testing
- Claude access cut over bioweapons concerns
- Every operation in Anthropic’s threat report was disrupted
- Fourth hacking case tied to an early Claude version
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

