Japan's government advisory council on AI decided on June 26, 2026, to begin a full-scale review of legal frameworks to address high-performance AI models such as Anthropic's Claude Mythos. It will establish a new working group (WG) to organize the issues for amending the AI Act or enacting new legislation.
July cabinet decision target · Japan AI Expert Panel
Japan moves to revise its AI laws as "Claude Mythos 5" raises cyber-threat alarms
A working group will launch an "active and continuous review" of legal frameworks, including the AI Act — triggered by Anthropic's world-leading, vulnerability-finding model and the risk of its misuse in cyberattacks.
Dec 2025
First AI Basic Plan adopted
~6 mo
Until first revision, prompted by rapid AI advance
Jul 2026
Target for cabinet decision on revised plan
SWE-bench Pro — software engineering capability
The trigger: Mythos 5 outperforms earlier models, with "world-leading" security ability.
Mythos 5 — restricted
"World-leading" cybersecurity ability; recognized as a cyber threat . Safeguards partly removed, offered only to defenders via "Project Glasswing." General release undecided.
Fable 5 — general use
Conservative safeguards. Broadly available on the API, AWS Bedrock, Google Cloud and Microsoft Foundry.
1M
context window (tokens)
$10 / $50
input / output per M tokens — under half the Preview
64.5%
Humanity's Last Exam (with tools)
Japan's regulatory direction
"Balancing innovation with risk response" — translating regulatory design into concrete measures.
Mandatory information provision by operators on performance and risk
AISI to evaluate cyber performance of domestic and overseas models
Stronger cooperation with foreign governments and operators
Continuous review of legal frameworks, including the AI Act
⚠ Sudden loss-of-access risk
A U.S. order restricted foreign access on national-security grounds, and Anthropic halted provision to all customers — illustrating how abruptly companies and developers can lose access. Officials had earlier voiced concern over "unknown unknowns."
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…