There is a rumor circulating that Anthropic will release Mythos to the public as early as this week. The company has not confirmed any specific date. What it has confirmed is that Mythos-class models will reach general availability "in the coming weeks," a phrase that appeared in the Opus 4.8 announcement on May 28. Polymarket gives a July launch 65% odds. Whatever the actual timeline, the crypto industry is running out of runway.
The Zcash Incident Was a Dress Rehearsal
On May 29, security researcher Taylor Hornby pointed Opus 4.8 at Zcash's Orchard shielded pool. Within 24 hours of the model's public release, he found a critical vulnerability that had survived four years of expert human review. Hornby built a working exploit in a local test environment. He disclosed it responsibly. Zcash teams deployed an emergency patch by June 1.
ZEC token price crashed nearly 50% in 48 hours. Arthur Hayes publicly exited his entire position. The market repriced that uncertainty immediately.
But here is the part that should concern anyone holding tokens outside the Glasswing perimeter: Opus 4.8 is the public model. Zcash's Orchard bug was found with a tool anyone can access for $5 per million input tokens. Mythos is materially more capable. Anthropic's own disclosures show Mythos scoring 97.6% on the 2026 USA Mathematical Olympiad versus Opus 4.6's 42.3%. On expert-level hacking tasks, the UK AI Security Institute reported a 73% success rate on challenges no AI model could complete at all until April 2025. During internal testing, the model discovered and exploited zero-day vulnerabilities in every major operating system and web browser.
Who Has the Head Start
Project Glasswing launched April 7 with roughly 50 partners. That list includes Amazon, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, Palo Alto Networks, and JPMorgan Chase. Anthropic has since expanded access to approximately 200 organizations across more than 15 countries, covering healthcare, power, water, communications, and hardware. The company has committed $100 million in usage credits and $4 million in donations to open-source security organizations.
These partners have been scanning their codebases with Mythos for two months. They have identified more than 10,000 high- or critical-severity vulnerabilities across systemically important software. Of the findings independently reviewed, over 90% proved valid. Cloudflare published a detailed account of its experience with the model, describing it as "a different kind of tool doing a different kind of work."
The crypto sector's representation in Glasswing is thin. JPMorgan is there. A handful of exchanges may have access through the second expansion wave, though Anthropic has not named specific crypto participants. What we know with certainty is that the vast majority of DeFi protocols, layer-1 chains, and infrastructure providers are not inside the program. They have not had two months to audit their code with Mythos. They have not even had two weeks.
The Asymmetry Problem
OpenZeppelin CEO Manuel Aráoz put it bluntly in late May: he now considers all of DeFi unsafe. His reasoning is structural. Defenders need to find and fix every bug. Attackers need to find and exploit one. AI tips that asymmetry further in the attacker's favor because it collapses the time between vulnerability discovery and working exploit from weeks to hours.
The numbers in 2026 support his concern. DeFi has already lost approximately $840 million to exploits through the first five months of the year. April alone accounted for over $600 million, driven by the Kelp DAO and Drift Protocol hacks. Annualized, 2026 is tracking toward $2 billion in DeFi theft. CertiK's CEO told CoinDesk that well-funded attackers can spend $10,000 to $20,000 keeping advanced AI engines running continuous vulnerability scans against a protocol for weeks. Protocol defenders operate under far tighter budgets.
This dynamic predates Mythos. But Mythos accelerates it. A general release means anyone with a subscription can point state-of-the-art vulnerability discovery at any codebase, anywhere, without restriction. The zombie chains from 2017 that never updated their cryptographic assumptions are obvious targets. So are the mid-tier DeFi protocols that raised $20 million in 2021, shipped a product, got one audit, and have not revisited their code since.
What Zcash Learned That Others Have Not
Zcash's response to the Orchard disclosure was competent. The patch shipped in days. Shielded Labs is now proposing a network upgrade with turnstile accounting on coins exiting Orchard, which would let anyone verify supply integrity going forward. The protocol took the hit, disclosed transparently, and is building toward a more defensible architecture.
The Zcash team also demonstrated something else: they were already using AI for defensive auditing. Hornby was hired specifically to find protocol vulnerabilities before malicious actors could. He found the Orchard bug within 24 hours of Opus 4.8's release because he was running the model against Zcash code as soon as it became available. That is the posture that survives in a Mythos world.
Most protocols are not in that posture. They are still operating on annual compliance audits and the assumption that if a codebase has been live for years without incident, it is probably secure. The Orchard bug disproved that assumption. It had been live since May 2022. It took four years and an AI model to find it.
The Window Is Closing
Anthropic has been explicit about the timeline. The company expects other AI labs to ship Mythos-class models within 6 to 12 months, potentially without safeguards. Its rationale for Glasswing was to give defenders a head start before that happens. But the head start is finite, and crypto is not evenly distributed inside the defended perimeter.
There are teams that have quietly secured Glasswing access through the second expansion wave. There are others exploring emerging security frameworks that incorporate AI-driven continuous monitoring. And there are protocols that have done neither, whose treasuries hold hundreds of millions in user funds, whose smart contracts were last audited in 2023.
When Mythos-class models reach general availability, the distinction between these categories will become measurable in TVL and in losses. Teams like Zcash have a headstart on tackling this new category of vulnerability exposure; how will other protocols fare?
