AI Daily Brief — 7 April 2026

Anthropic announces Claude Mythos Preview, a frontier model the company describes as a full capability tier above Opus 4.7. Mythos lands paired with Project Glasswing — an invitation-only program with 12 founding organisations and roughly 40 vetted critical-infrastructure operators.

Top stories

• Claude Mythos Preview. Anthropic’s first frontier-class system since Opus 4.7. Mythos is not generally available at launch; access runs only through Project Glasswing partners. Pricing for partners: $25 per million input tokens, $125 per million output — five times the cost of Opus 4.7. via Anthropic
• Cybersecurity capabilities & CVE-2026-4747. The Red team’s assessment reports that Mythos Preview fully autonomously identified and exploited a 17-year-old remote-code-execution vulnerability in FreeBSD that allows root on any machine running NFS, triaged as CVE-2026-4747. Across Glasswing partners, Mythos surfaced more than 10,000 high- or critical-severity vulnerabilities. via Anthropic Red
• Project Glasswing partners. AWS, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks form the founding consortium. via InfoQ

Quiet corners

The frontier flagship table updates to include Mythos as a tier above Opus 4.7 — for the dozen orgs that can use it.

By the numbers

• 10,000+ high/critical vulnerabilities surfaced across Glasswing partners
• $25 / $125 per M input/output tokens for Glasswing partners
• 5× the price of Opus 4.7
• 12 founding partners + ~40 vetted critical-infra operators

Compiled by AI Feed’s editor from publicly reported announcements on 7 April 2026.