Anthropic’s newest AI model can find security flaws that human researchers and automated tools have missed for decades. For DeFi protocols holding billions in user funds, that changes the threat calculus overnight.
On April 7, Anthropic launched Project Glasswing, a cybersecurity initiative built around Claude Mythos Preview, a frontier AI model the company describes as its most capable for coding and agentic tasks. Mythos has already identified thousands of zero-day vulnerabilities across major operating systems, browsers, and cryptographic libraries. Anthropic is restricting access to defensive security researchers only, but the implications for crypto are hard to ignore.
A zero-day vulnerability is a software flaw unknown to the vendor, meaning no patch exists when it is discovered. Attackers who find zero-days first can exploit systems before anyone knows the weakness is there.
Key takeaways
- Claude Mythos Preview found a 27-year-old bug in OpenBSD for under $50 in compute and a 16-year-old flaw in FFmpeg that five million automated scans missed.
- The model discovered weaknesses in TLS, AES-GCM, and SSH, cryptographic protocols that underpin most DeFi smart contract infrastructure.
- Anthropic committed $100 million in usage credits and partnered with 12 companies including AWS, Microsoft, Google, and JPMorgan Chase to fix vulnerabilities before attackers find them.
- DeFi protocols may need to move beyond friction-based defenses like multisig and timelocks toward formal verification and cryptographic proofs.
What Mythos found and why it matters for crypto
The raw numbers are striking. Mythos uncovered a 27-year-old memory corruption bug in OpenBSD for under $50 in compute costs. It flagged a 16-year-old vulnerability in FFmpeg, a media processing library used across the internet, that had survived five million scans by existing automated tools. In one test, the model turned a publicly known Linux vulnerability into a full working exploit in under 24 hours for less than $2,000, a task that would take a skilled human researcher weeks.
The model also autonomously chained four separate vulnerabilities into a single browser exploit that escaped both the renderer and operating system sandboxes. Palo Alto Networks, one of the 12 launch partners, confirmed that Mythos “identified complex vulnerabilities that prior-generation models missed entirely.”
For DeFi, the critical finding is that Mythos discovered weaknesses in TLS, AES-GCM, and SSH. These are not obscure libraries. They are the cryptographic backbone of wallet communications, node-to-node messaging, key management systems, and the infrastructure that most DeFi protocols rely on daily.
The cost equation that should concern DeFi builders
DeFi code is public by design. Every smart contract on Ethereum, Solana, or any other chain sits in the open, readable by anyone. Until now, the cost of finding zero-day exploits in that code required significant human expertise, time, and resources. Mythos collapses that cost to near zero at machine speed.
The $285 million Drift Protocol hack earlier this month, traced to North Korean operatives, demonstrated what happens when sophisticated attackers target DeFi infrastructure. An AI model that can autonomously catalog every weakness in a codebase makes the attacker’s job orders of magnitude easier.
Standard DeFi safeguards like multisig governance, timelocks, and third-party audit reports rely on friction. They slow attackers down. They do not stop an adversary that can scan entire codebases in minutes. As Cisco’s Senior Vice President and Chief Security Officer Anthony Grieco put it: “AI capabilities have crossed a threshold that changes the urgency required to protect critical infrastructure.”
Project Glasswing’s defensive play
Anthropic is not releasing Mythos to the public. The model is restricted to Project Glasswing’s 12 founding partners: AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. More than 40 additional organizations that maintain critical software infrastructure also have access.
Anthropic committed $100 million in usage credits and $4 million in donations to open-source security organizations. Discovered vulnerabilities follow a 90-day coordinated disclosure timeline, giving maintainers time to patch before details go public.
The pricing for participants is $25 per million input tokens and $125 per million output tokens. That is expensive compared to standard AI model access but cheap compared to the cost of a major DeFi exploit.
What comes next for DeFi security
The DeFi industry lost $3.4 billion to hacks in 2025, with the Bybit breach alone accounting for $1.4 billion. Q1 2026 showed improvement with an 89% drop in losses, but Drift’s $285 million exploit proved that single catastrophic events can still occur.
If Mythos-level capability reaches malicious actors, whether through model replication, jailbreaking, or independent development by other AI labs, the current security model for DeFi breaks down. Protocols will need to shift from friction-based defenses to hard barriers: formal verification of smart contracts, cryptographic proofs for state transitions, and architecture designed to remain safe even when individual components are compromised.
Ledger’s CTO warned that “AI is making cyberattacks faster, cheaper, and more scalable.” The window between Mythos existing as a defensive tool and similar capabilities appearing in the wild is the time DeFi has to adapt.
Frequently asked questions
What is Claude Mythos Preview and why is it restricted?
Claude Mythos Preview is Anthropic’s most capable AI model for coding and security research. It can autonomously discover zero-day vulnerabilities across operating systems, browsers, and cryptographic libraries at extremely low cost. Anthropic restricted access to defensive researchers through Project Glasswing because the same capabilities that find bugs could also be used to exploit them.
How does Mythos threaten DeFi protocols specifically?
DeFi smart contracts are publicly readable, giving AI models a complete view of the codebase to analyze. Mythos can scan entire protocols at machine speed for near-zero cost, potentially finding exploitable flaws faster than human auditors. It also found weaknesses in cryptographic libraries like TLS and AES-GCM that DeFi infrastructure depends on.
What should DeFi projects do to prepare?
Security researchers recommend moving beyond friction-based defenses like multisig and timelocks toward formal verification of smart contracts, cryptographic proofs for state transitions, and resilient architecture. Protocols should also consider engaging with AI-powered defensive auditing tools rather than relying solely on traditional code audits.
