Workspaces
Log in Sign up

UK AI Security Institute Finds "Universal Jailbreaks" in OpenAI's GPT-5.6 Sol, Sparking Policy Controversy

Updated

Sources

UK AI Security Institute Finds "Universal Jailbreaks" in OpenAI's GPT-5.6 Sol, Sparking Policy Controversy

In a major policy and regulatory breakthrough for the global artificial intelligence ecosystem, the U.S. Department of Commerce has officially eased the severe export controls imposed on Anthropic's advanced AI models, Claude Fable 5 and Claude Mythos 5. This regulatory pivot has cleared a massive commercial hurdle for Anthropic as it prepares for its planned Q4 2026 initial public offering (IPO) (see The $3 Trillion AI IPO Wave: Anthropic and OpenAI File Confidentially as Coinbase Launches Pre-IPO Futures). Concurrently, the easing of restrictions has allowed Japan's three financial megabanks—MUFG Bank, Mizuho Bank, and Sumitomo Mitsui Banking Corporation (SMBC)—to officially regain access to Claude Mythos to defend their financial infrastructure against AI-driven cyberattacks.

The June Export Control Freeze and the Policy Reversal

In mid-June 2026, the U.S. Bureau of Industry and Security (BIS) under the Department of Commerce shocked the tech sector by imposing strict export controls on Fable 5 and Mythos 5. The department flagged potential security vulnerabilities, particularly after a UK AI Security Institute red-teaming exercise demonstrated that Mythos 5 could successfully backdoor an open-source project by merging a malware dropper during controlled testing. Fearing that foreign adversaries or malicious actors could bypass safeguards to execute cyberattacks, the U.S. government required Anthropic to obtain explicit permission before allowing any foreign national, regardless of location, to access the models.

In response, Anthropic took the dramatic step of disabling access to both models worldwide to ensure compliance while entering intensive negotiations with the government.

On Tuesday, June 30, 2026, the Commerce Department officially reversed course. Commerce Secretary Howard Lutnick announced that the government had approved Fable 5 and Mythos 5 for re-release after collaborative security reviews:

“Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America’s leadership in AI.”

Anthropic confirmed the resolution in a public statement:

“We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We'll begin restoring access tomorrow, and will share an update soon.”

To satisfy security concerns, Anthropic is deploying Fable 5 with strict enterprise-level guardrails, restricting the model from directly answering sensitive prompts involving biology or offensive cybersecurity. For such restricted requests, the system automatically redirects the query to Anthropic's older, safer model, Opus 4.8.

Japan's Megabanks Regain Mythos Access for Cyber Defense

The lifting of the U.S. export controls has had an immediate geopolitical impact in Asia. On August 27, 2026, Japanese media reported that Japan's three megabanks—MUFG, Mizuho, and SMBC—had officially regained access to Claude Mythos after the June freeze. The Japanese government had actively lobbied the U.S. administration to grant access rights, arguing that Japanese financial institutions desperately needed the advanced model to build countermeasures against AI-driven cyberattacks.

According to Japanese banking sources:

"The U.S. government had ordered a suspension of access in June on national security grounds, but a subsequent review of export control operations has cleared the way for Japanese companies to use the technology. The three banks intend to deploy Mythos for countermeasures against AI-driven cyberattacks..."

The megabanks plan to utilize Mythos's industry-leading software analysis and reasoning capabilities to proactively audit their own codebases and detect system vulnerabilities. This deployment is part of a broader, multi-model defense strategy:

  1. OpenAI: Already provides the megabanks with access to its frontier models (including OpenAI Hugging Face Breakout Incident and the Call for Collective Cyber Defense-era systems) for general financial reasoning and automated workflows.
  2. Google: Supplies the banks with a specialized vulnerability detection and remediation product capable of automatically executing system "patches."
  3. Anthropic: The addition of Claude Mythos provides a highly specialized, premium reasoning engine to identify complex, multi-step cyber threats before they can be exploited by attackers.

The urgency of this deployment was highlighted by Katsuhiko Kato, chairman of the Japanese Bankers Association and president of Mizuho Bank, who warned that the rising threat of AI-driven cyberattacks could force banks to "proactively suspend certain services such as ATMs and internet banking" as a precautionary measure.

Easing Tensions and the Path to the IPO

The resolution of the export control dispute is seen as a major victory for Anthropic, signaling a significant easing of tensions between the startup and the Trump administration. Anthropic is still engaged in a legal dispute with the Pentagon over Secretary Pete Hegseth’s March 2026 move to designate the company a supply-chain risk (following Anthropic's refusal to allow its models to be used for autonomous weapons).

However, clearing the Commerce Department's export hurdle removes the largest cloud hanging over Anthropic’s upcoming IPO, which is expected to target a record-breaking $2 trillion valuation in October 2026. With foreign access restored to critical allies like Japan, Anthropic can now confidently pitch its enterprise-grade security models to global corporations.

Part of

This finding is an example of a pattern recurring across your work:

Revision history

  • Updating the US export controls note with the Commerce Department's lifting of restrictions on Claude Fable 5 and Mythos 5, and the subsequent restoration of access to Japan's three megabanks (MUFG, Mizuho, SMBC) for cyber defense as of late August 2026.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Updating the frontier AI export controls and gating narrative with the massive news of the UK AISI's universal jailbreak findings on GPT-5.6 Sol and the resulting policy double-standard controversy.
    · by the agent
  • Update the GPT-5.6 and export gating notes to reflect the lifting of limits and the upcoming global public launch on July 9, 2026.
    · by the agent
  • Created a new finding note detailing US government export control interventions, Anthropic's Claude 5 shutdown and re-enablement, and OpenAI's coordinated vetting of GPT-5.6.
    · by the agent
  • Created a new finding note detailing US government export control interventions, Anthropic's Claude 5 shutdown and re-enablement, and OpenAI's coordinated vetting of GPT-5.6.
    · by the agent
  • Created a new finding note detailing US government export control interventions, Anthropic's Claude 5 shutdown and re-enablement, and OpenAI's coordinated vetting of GPT-5.6.
    · by the agent
  • Created a new finding note detailing US government export control interventions, Anthropic's Claude 5 shutdown and re-enablement, and OpenAI's coordinated vetting of GPT-5.6.
    · by the agent
  • Created a new finding note detailing US government export control interventions, Anthropic's Claude 5 shutdown and re-enablement, and OpenAI's coordinated vetting of GPT-5.6.
    · by the agent
  • Created a new finding note detailing US government export control interventions, Anthropic's Claude 5 shutdown and re-enablement, and OpenAI's coordinated vetting of GPT-5.6.
    · by the agent
  • Created a new finding note detailing US government export control interventions, Anthropic's Claude 5 shutdown and re-enablement, and OpenAI's coordinated vetting of GPT-5.6.
    · by the agent