Anthropic, the AI startup behind the Claude model, has launched a bold initiative called Project Glasswing that brings together some of the world’s biggest tech companies to tackle the rising cybersecurity risks posed by advanced artificial intelligence. The stakes are high, with vulnerabilities found in nearly every major operating system and web browser.
A New Front in Cybersecurity: AI vs. AI
Diesel prices hitting $5 a gallon in parts of Texas last week might catch your attention, but the real news is Anthropic’s launch of Project Glasswing, a coalition of major tech companies working to fight AI-powered cyberattacks. Microsoft, Apple, Google, Amazon Web Services, Nvidia, and over 40 other companies from cybersecurity, infrastructure, and finance sectors have signed on. They’re getting early access to Anthropic’s unreleased AI model, Claude Mythos Preview, to test and shore up defenses.
Here’s the thing — Mythos Preview wasn’t designed specifically for cybersecurity. It was built to be good at code. But that skill set gave it the side effect of being able to spot security flaws and develop potential exploits. That’s both a blessing and a curse. While it helps defenders find vulnerabilities faster, it also arms bad actors with new tools to launch sophisticated attacks that used to be too complicated or costly.
Anthropic’s CEO, Dario Amodei, put it bluntly: more powerful AI models are coming from multiple sources. The industry needs a plan to manage the fallout.
The goal with Project Glasswing is to get ahead of the problem by letting trusted partners safely use Claude Mythos Preview to identify weak spots and fix them before attackers do.
How Project Glasswing Works
Participants in Project Glasswing will use Mythos Preview to scan their own systems and software for security gaps. This model has already uncovered thousands of exploitable vulnerabilities, cutting across every major operating system and web browser. It can simulate attacks, generate proofs of concept, and even hunt for system misconfigurations that humans might miss.
More advanced than typical vulnerability scanners, Mythos Preview can conduct penetration testing and assess endpoint security without needing access to source code. That means it can analyze compiled binaries and other hard-to-inspect components. Anthropic is rolling out the model in phases, starting with this industry collaboration before any broader public release.
Logan Graham, Anthropic’s frontier red team lead, made it clear the effort isn’t about promoting the model or the company. Instead, it’s about preparing for a future where AI capabilities like this become widely available. That future could arrive within 6, 12, or 24 months. When it does, many of today’s security assumptions might break down.
Big Tech and Financial Firms Unite
The list of Project Glasswing partners reads like a who’s who of tech and cybersecurity: Microsoft, Apple, Google, Amazon Web Services, Nvidia, Broadcom, Cisco, CrowdStrike, Palo Alto Networks, and even financial giant JPMorgan Chase. The Linux Foundation is also involved, representing open-source software interests. Together, they're tackling the challenge from multiple angles — software platforms, cloud infrastructure, hardware, and financial systems.
Anthropic has committed up to $100 million in usage credits and $4 million in donations to open-source security groups. This funding aims to spur broader development of defenses against AI-driven attacks. The startup is also in ongoing discussions with the U.S. Government about how to handle the risks and benefits of these models.
Still, the move comes amid controversy. Earlier this year, Anthropic refused Pentagon requests to remove AI safety guardrails on its Claude service, leading to a Department of Defense designation flagging Anthropic as a supply chain risk. And Claude itself was reportedly exploited by hackers to target government agencies in Mexico. So, while Project Glasswing signals cooperation, the road ahead is complicated.
The Rising Tide of AI Cyberattacks
Cybersecurity experts have been warning for months about the rise of AI-powered attacks. At the 2024 RSA cybersecurity conference in San Francisco, industry leaders debated whether traditional security tools can keep pace with these new threats. Studies back up the concern: a recent survey of 1,000 executives by IBM and Palo Alto Networks found that 67 percent had been targeted by AI-fueled attacks in the past year.
That said, aI models like Claude Mythos Preview accelerate the classic cat-and-mouse game between hackers and defenders. But the speed and scale have jumped dramatically. Automated tools can now discover vulnerabilities and craft exploits in minutes rather than weeks or months. That makes it easier for criminals to target critical infrastructure, financial systems, and personal data.
Anthropic’s approach of sharing its model with vetted industry partners aims to flip AI’s double-edged sword into a force for good. By using AI to hunt down and patch flaws, defenders can stay one step ahead. But it’s a race against time as more powerful models emerge from other players.
What’s Next for AI and Cybersecurity?
The stakes are incredibly high. As AI models grow smarter, the potential for widespread damage from automated cyberattacks increases. The collaboration around Project Glasswing is a step toward building defenses before disaster strikes.
Anthropic aims to safely deploy models like Mythos on a large scale. That means not just internal use but potentially broader access under controlled conditions. How that balance between openness and security plays out will shape the future of AI in cybersecurity.
Meanwhile, companies outside the initial coalition are watching closely. Will they follow suit or develop competing approaches? How will regulators respond to this new frontier? The answers could reshape trust in technology and global digital security for years to come.
For now, Anthropic and its partners are betting that cooperation and transparency offer the best chance to keep AI from hacking everything.
Anthropic's Project Glasswing brings together major industry players to confront AI’s cybersecurity challenges head-on, aiming to turn potential threats into defensive advantages before more powerful models hit the mainstream.
