Anthropic executives traveled to Washington, DC on June 15, 2026 to meet directly with White House and Commerce Department officials, but the two sides remain at odds over how to assess the risks of the now-suspended new model Claude Fable 5.
June 16, 2026 · Anthropic × White House
Still at Odds Over "Claude Fable 5"
Anthropic executives flew to Washington to argue that national-security fears over its new model are "overblown" — but after Monday's high-level talks, the two sides remain divided and the export-style access restrictions stay in place.
2
models unveiled together on June 9 — same foundation, different rules
1M
token context window; up to 128K max output
1–3
point benchmark gap between Fable 5 and unrestricted Mythos 5
Two builds, one foundation
How close the restricted and unrestricted versions sit — benchmark gap stays within 1–3 points.
Fable 5
safety-strengthened
Mythos 5
unrestricted
The government's fear: Fable 5's guardrails could be removed — unlocking Mythos-class dangerous capabilities through the "safe" version.
The standoff
NSA warns guardrails could be bypassed
→
Access restricted for foreign nationals — incl. Anthropic's own staff
→
June 16 D.C. talks — restrictions not lifted
Praised by developers
"Relentlessly proactive" agentic work
Finds and fixes dependency bugs on its own
SOTA-class on software, science, vision
Practical stumbling blocks
High cost — $10 in / $50 out per 1M tokens
Frequent refusals and guardrail activations
Usage limits hit in a single session
"You never know when the government might switch a model off."
The continued restrictions are spreading unease across AI labs — and some suggest the friction could accelerate a shift toward open models. The tug-of-war over security and access looks set to continue.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…