30초 요약
In a June 2026 video, Matthew Berman discusses Anthropic’s suspension of Fable 5 and Mythos 5 after a US government directive and argues that jailbreaks and cyber capabilities are industry-wide issues. Anthropic subsequently said export controls were lifted and access was restored, with Mythos availability still limited to selected partners.
- 1A claimed jailbreakBerman cites a post claiming a jailbreak soon after Fable 5 launched and questions the implications of shutting down a model on which customers depend.
- 2Anthropic’s accountAnthropic said the government had not specified its concern and that the technique it reviewed found a few previously known, relatively minor vulnerabilities.
- 3Cyber capability and comparisonBerman worries most about the discovery of unknown software vulnerabilities, but argues competing models may have similar capabilities.
핵심 정리
- Anthropic said a June 12 directive required it to restrict foreign-national access, so it temporarily disabled both models for all customers.
- Anthropic described the reported bypass as narrow and said its demonstration involved known vulnerabilities; Berman argues safeguards cannot eliminate every jailbreak.
- Berman questions why similar cyber capabilities in other models did not receive the same treatment.
- Anthropic later announced that the export controls had been lifted and access restored.
챕터별로 다듬은 전문 읽기
A jailbreak and a shutdown
Berman points to an alleged early Fable 5 jailbreak and the disruption caused by taking models offline. Anthropic said the government’s directive led it to suspend both models.
Pliny the Liberator posted a claimed Fable 5 jailbreak within hours of release. I see such attempts as a recurring feature of new model launches, so the decision to shut the model down is striking, especially for companies that have come to rely on it. A claim of a jailbreak, however, does not establish a universal bypass or explain which report triggered the directive.
What Anthropic said
He reads the company’s description of the demonstration and its limits. He argues that jailbreak risk can be reduced but not entirely eliminated.
Anthropic said the government believed it had learned of a method of bypassing Fable 5. The company said it reviewed a demonstration involving a small number of previously known, relatively simple vulnerabilities that other publicly available models could also identify without a bypass. Its June 12 statement said the government’s letter had not specified the national security concern. [Anthropic’s statement](https://www.anthropic.com/news/fable-mythos-access) described the directive as restricting foreign-national access, which prompted it to suspend both Fable and Mythos for all customers.
My view is that safeguards can minimize jailbreaks but cannot guarantee their complete prevention. I also argue that sensitive topics can sometimes be elicited from models, though capability and risk depend on the particular model, prompt and context; a general claim about every model should not be treated as proof of a specific bypass.
Cyber risks across the industry
Berman distinguishes routine information from the risk of discovering unknown software flaws. He questions the unequal treatment of models he considers comparable.
The more serious cyber risk, in my view, is a model finding software vulnerabilities that nobody else knows about. I question whether Fable and Mythos are sufficiently ahead of GPT-5.5 in that respect to justify treating them differently. That comparison is my assessment, not a comprehensive independent evaluation of the models.
I think Anthropic’s own warnings about dangerous cyber capabilities now shape the scrutiny it faces, while OpenAI continues to offer a model I regard as nearly as capable. My prediction that future models will be stronger and susceptible to jailbreaks is speculation. In a later [June 30 update](https://www.anthropic.com/news/redeploying-fable-5), Anthropic said the export controls were lifted and Fable access would resume globally on July 1; Mythos access was restored to a limited set of organizations while the company sought to expand it.
같은 채널의 다른 글
34:21EN
Matthew Berman· 2026년 10월 27일Matthew Berman Examines Anthropic’s IPO Filing, AI Growth and GovernanceMatthew Berman, Sam Altman · Forward FutureThe presenter reviews the S-1’s financial metrics, compute strategy and voting structure while questioning the tension between rapid AI development and safety.
0:26ENMatthew Berman· 2026년 10월 25일Sam Altman on Human Values and Problem-Solving in an AI WorldSam Altman · OpenAIIn a short interview excerpt, Altman considers the enjoyment of solving hard problems and the human role in interpreting results.
0:36ENMatthew Berman· 2026년 10월 25일Kimi K3.1 and a Rubik’s Cube Solver DemonstrationMatthew Berman · Forward FutureMatthew Berman shows a scrambled virtual cube being solved and gives the Kimi K3.1 test a favorable assessment.
0:21ENMatthew Berman· 2026년 10월 24일An AI-Generated Kimi Benchmark Chart Used Invented ScoresMatthew Berman · Forward FutureMatthew Berman asks for a graph of Kimi K3.1 Pro benchmarks, then discovers that the displayed figures are illustrative placeholders.
25:50ENMatthew Berman· 2026년 10월 24일Kimi K3.1 Pro Tested: Strong 3D Demos, Uneven Results and Competitive PricingMatthew Berman · Forward FutureMatthew Berman tests the open-weight model on coding, 3D rendering and visual explanations, then compares benchmark results and prices.
32:20ENMatthew Berman· 2026년 10월 23일Who Is Local AI For? The Costs, Trade-offs and a Hybrid FutureMatthew Berman · Forward FutureMatthew Berman weighs data control and independence against hardware cost, setup complexity and inference speed.