Breaking

Anthropic CEO Meets US Officials Over New Mythos Model

June 15, 2026 at 02:31 EDT

Foundation Models
Policy & Regulation
AI Agents

Anthropic CEO Dario Amodei met White House officials on April 17, 2026, to discuss the responsible use of the company's new AI model "Mythos" and possible areas of cooperation. Attendees included White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent, and both sides described the meeting as "productive and constructive." Details

April 17, 2026 · Anthropic × White House

A Thaw at the White House Over Anthropic's Most Powerful Model Yet

CEO Dario Amodei met Trump officials to discuss the responsible deployment of new frontier model "Mythos" and cybersecurity cooperation — both sides called the talks "productive and constructive," easing an ongoing Pentagon contract dispute.

Mythos vs. Claude Opus 4.6 — benchmark leap

Where existing benchmarks near saturation, the new model widens the gap most on harder coding and multimodal tasks.

SWE-bench Pro

53.4%

Opus 4.6

77.8%

Mythos

SWE-bench Verified

80.8%

Opus 4.6

93.9%

Mythos

Multimodal eval

27.1%

Opus 4.6

59.0%

Mythos

2.0×

Multimodal score vs. previous model (59.0% vs 27.1%)

$25 / $125

Per M input / output tokens — invite-only access

Apr 7

Mythos previewed — the turning point that eased tensions

"Too powerful for general release"

Anthropic limited Mythos to Project Glasswing — invitation- and approval-only — citing strong cyber capabilities that lead significantly in finding and exploiting zero-day vulnerabilities.

Select partners only Security testing focus API · Bedrock · Vertex

How the dispute unfolded

Policy bans autonomous weapons & mass surveillance

→

Pentagon demands "all lawful uses" — talks end

→

Labeled "supply chain risk," gov-wide phaseout ordered

→

Partial injunction (Mar 2026) → April thaw

White House view

Acknowledges Mythos's value for cyber defense; talks called "productive and constructive," with both sides willing to continue dialogue.

Lingering concerns

Pentagon still presses for "lawful uses" and is dissatisfied with guardrails; researchers warn of autonomous exploit creation. No concrete deal announced.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.