Reuters · 2026-05-12 · major
Anthropic Mythos Forces US Megabanks Into Days-Not-Weeks Cyber Patching — JPMorgan, Goldman, Citi, BofA, Morgan Stanley Are Already Inside the Glasswing Preview
Reuters reports that JPMorgan, Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley are using Anthropic's Mythos preview to scan their stacks; banks are now patching hundreds-to-thousands of low-to-moderate vulnerabilities in days rather than weeks at $25/$125 per million tokens.
Five of the largest US banks already have Mythos access and are racing to plug 'several hundred to thousands' of vulnerabilities the model surfaced.
Key specs
| Input price per million | $25 |
|---|---|
| Output price per million | $125 |
| Mythos to opus 4 7 multiplier | 5x |
| Named banks | 5 |
| Glasswing credits | $100M |
What is it?
An enterprise adoption story rather than a product launch: Anthropic's Project Glasswing preview is in active use at JPMorgan Chase (the original publicly-named launch partner), Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley. Mythos is the restricted Anthropic model behind Glasswing, tuned for security work and offered only to vetted defenders. Those banks are sharing summaries of what Mythos finds with smaller lenders who do not have direct access.
How does it work?
Mythos runs against bank codebases and infrastructure and surfaces chains of lower-risk weaknesses that combine into high-risk attack paths — the same pattern Microsoft's MDASH harness used to find sixteen Windows flaws last week. Pricing is steep by Anthropic standards: $25 per million input tokens and $125 per million output tokens, five times the Opus 4.7 rate. Anthropic is offsetting cost by handing Glasswing partners $100M in usage credits. Banks say the volume of findings — several hundred to several thousand low-to-moderate vulnerabilities per institution — is forcing them to compress patch cycles from weeks to days and accelerate retirement of out-of-support software, with possible customer-facing maintenance impact.
Why does it matter?
This is the first concrete proof that restricted frontier defender models change real-world enterprise workflows, not just lab benchmarks. CrowdStrike's Adam Meyers reacted with 'oh boy' when briefed; Incedo's Nitin Seth told Reuters cyber risk is now 'moving at machine speed' while most bank defense still operates at human speed. It also sharpens the policy debate: White House officials are openly weighing pre-release vetting for frontier models, with Mythos cited as the catalyst.
Who is it for?
Bank CISOs, financial-sector security teams, AI policy watchers
Try it
https://www.anthropic.com/news