Anthropic warns its latest AI model could supercharge cyberattacks, refuses to release ‘Claude Mythos’. Rushing to arm defenders first.
Summary:
- Anthropic restricts release of new AI model over cyberattack risks
- Model given to major firms to strengthen defenses first
- AI agents could dramatically accelerate hacking speed and scale
- Concerns growing over widening gap between attackers and defenders
- Model reportedly finding vulnerabilities at unprecedented rates
- US officials briefed, highlighting national security implications
- Controlled rollout aims to prepare before broader AI proliferation
Anthropic is moving cautiously with its latest artificial intelligence model, warning that its capabilities could significantly accelerate cyberattacks if widely released, even as it rolls out the technology to major corporations in a controlled effort to strengthen global cyber defenses.
The model, known as “Claude Mythos Preview,” is being made available to a select group of leading firms; including Amazon, Apple, Microsoft and JPMorgan Chase, as well as cybersecurity specialists and infrastructure providers. The aim is to identify vulnerabilities in widely used systems before malicious actors can exploit them.
Anthropic has deliberately held back from a public release, citing concerns that the model’s offensive capabilities could be misused by hackers or state actors. Internally and among policymakers, there is growing unease that tools like Mythos represent a step-change in cyber risk, enabling attacks to be executed at speeds and scales far beyond traditional human limitations.
Experts say AI-driven agents could autonomously scan for weaknesses across vast systems and exploit them continuously, effectively compressing what once took teams of hackers days or weeks into minutes. That dynamic raises the prospect of a widening imbalance, where defensive capabilities struggle to keep pace with rapidly evolving offensive tools.
Anthropic has already briefed senior US government officials on the model’s capabilities, underscoring the national security implications. The company is also supporting official testing, highlighting the degree to which advanced AI is now being treated as both a commercial technology and a strategic asset.
Early indications suggest the model is highly effective. Anthropic claims Mythos has identified thousands of previously unknown software vulnerabilities in recent weeks—far exceeding the rate of traditional human-led discovery. While those figures remain unverified, they point to the scale at which AI could reshape the cybersecurity landscape.
The core risk lies in accessibility. While Anthropic is attempting to get ahead of potential threats by arming defenders first, the broader trajectory is clear: increasingly powerful AI systems are likely to become more widely available over time. If that occurs without adequate safeguards, the gap between attackers and defenders could widen materially, raising the probability of more frequent, faster, and potentially more severe cyber incidents.
For now, the controlled rollout reflects a race against time—strengthening digital defenses before the same capabilities inevitably proliferate more broadly.
AI is back as the main threat to civilization
