Anthropic Teams Up With Its Rivals to Keep AI From Hacking Everything

Anthropic Teams Up With Its Rivals to Keep AI From Hacking Everything

Summary

Anthropic has launched the private “Claude Mythos Preview” model and convened an industry consortium called Project Glasswing to examine the cybersecurity implications of more capable AI. More than 45 organisations — including Microsoft, Apple, Google, AWS, Cisco, Nvidia and the Linux Foundation — will get controlled access to the model so they can test it on their own systems, discover vulnerabilities, and develop mitigations before capabilities become broadly available.

Key Points

1. Project Glasswing is a cross‑industry coalition organised by Anthropic to probe and defend against AI‑driven cyber threats.
2. The Claude Mythos Preview model, not yet publicly released, is particularly good at code and therefore effective at finding vulnerabilities, creating exploit chains and assessing binaries without source code access.
3. Anthropic says Mythos Preview has already surfaced thousands of critical vulnerabilities, including long‑standing bugs missed by human reviewers.
4. The launch uses a staggered, coordinated disclosure approach to give platform and infrastructure providers time to patch issues before wider exposure.
5. Major tech firms and infrastructure operators are participating to identify risks early and augment existing security tools and practices.
6. Anthropic warns these capabilities will spread in months, forcing a rethink of current security paradigms and assumptions.

Content Summary

The article explains that Anthropic’s Mythos Preview is a step change in AI’s ability to analyse and manipulate code, which carries both defensive and offensive implications. Rather than releasing the model publicly, Anthropic is providing private access to a broad set of industry players so they can run the model against their systems and find exploitable weaknesses. The goal is to accelerate defensive work — patching, mitigation and new tooling — while acknowledging the risk that the same capabilities could empower attackers if released without safeguards.

Anthropic frames Project Glasswing as the start of a much larger effort: the company and its partners want to catalogue the key questions, surface systemic problems the model exposes, and scale defensive responses. Participants describe the collaboration as urgent and unprecedented because AI is changing the scale and pace of vulnerability discovery.

Context and Relevance

This matters because AI that can read, test and propose changes to code changes the software‑security landscape. For security teams, CISOs, developers, cloud providers and regulators, the article highlights an immediate challenge: how to use powerful models to improve security without accelerating attackers. It ties into broader trends in AI-assisted development, supply‑chain risk, and the arms race between defensive and offensive cyber capabilities.

Why should I read this?

Look — if you care about keeping systems running and data safe, this is one to skim properly. Anthropic’s move shows the tech giants are treating advanced code‑capable models as a genuine threat/opportunity. The piece tells you who’s involved, why they’re worried, and what they’re doing about it — useful if you make security decisions or build software.

Author style

Punchy: this isn’t a dry policy memo. The reporting flags an urgent industry pivot — Anthropic isn’t just debuting a model, it’s nudging the whole security ecosystem to scramble now rather than later. If you’re in security or run critical systems, the details here are worth a closer read.

Source

Source: https://www.wired.com/story/anthropic-mythos-preview-project-glasswing/