After a protracted dialogue with the Trump administration, Anthropic confirmed on Wednesday that the Department of Commerce has lifted export controls on its Claude Fable 5 model. The decision clears the way for the company to restore global access to the AI system, which had been sidelined in early June after officials raised concerns about potential jailbreaks.
In a post on X, Anthropic said it will begin restoring access the next day and will soon share an update on progress. The firm also indicated that it plans to re‑enable Fable 5 on Amazon Web Services, Google Cloud, and Microsoft Foundry, though it did not set a firm timeline for the cloud rollout.
Anthropic’s blog post, published Tuesday evening, detailed the steps the company took to address the jailbreak that triggered the export control directive. An Amazon‑research team had flagged a technique that could bypass the model’s safeguards. In response, Anthropic trained an improved safety classifier that blocks the exploit in more than 99 percent of cases. When a request is blocked, users are redirected to the older Opus 4.8 model, and they receive a notification about the action.
New safety measures and government partnership
The company emphasized that the updated classifier represents a significant upgrade to its defensive toolkit. It also announced a suite of initiatives aimed at tighter collaboration with the U.S. government. Anthropic will provide pre‑release access to its models for national‑security‑related evaluations, allowing government partners to run independent tests before broader deployment. During these testing periods, the agency will have direct access to Anthropic’s technical staff.
Beyond testing, Anthropic pledged rapid information sharing whenever a major jailbreak or misuse pattern emerges. The firm is joining forces with other leading AI labs to craft a voluntary, industry‑wide security and evaluation standard. To support these efforts, Anthropic will allocate dedicated compute resources, stand up specialized teams for government priorities, and make its safety‑and‑red‑team expertise available for joint research.
Anthropic also announced a 24/7 monitoring team that will track key jailbreak submission channels. A forthcoming HackerOne program will let external researchers submit potential jailbreaks for Fable 5, expanding the model’s security oversight.
While the export control lift restores access for most users, the Trump administration has kept Mythos 5, Anthropic’s more advanced model, limited to a pre‑approved list of organizations. Non‑U.S. members of those organizations and Anthropic’s foreign‑national employees may now regain access, but broader rollout remains constrained.
Anthropic’s leadership framed the development as a step toward normalizing AI deployment amid heightened regulatory scrutiny. The company, which is preparing for an initial public offering, said it will continue to coordinate with the government to expand access to both domestic and international partners.
Industry observers note that the timing coincides with OpenAI’s launch of GPT‑5.6, which also entered the market under a staggered rollout governed by similar export‑control rules. Both cases highlight the growing tension between rapid AI innovation and national‑security concerns.
Anthropic concluded its blog post with a disclaimer: no AI model can be made fully impervious to jailbreaks. The firm expects ongoing discovery of minor and more serious exploits, and it will keep its safety researchers engaged in red‑team activities.
Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.