U.S. Commerce Department Revises Export Controls on Anthropic’s AI Models
On Tuesday, Anthropic announced that the U.S. Commerce Department has rescinded the export controls that had previously compelled the company to deactivate its two premier artificial intelligence (AI) models, Fable 5 and Mythos 5, for users around the globe, including India.
This policy shift concludes a saga that commenced on June 12, when the Trump administration mandated that Anthropic restrict access to these models for all foreign nationals on the grounds of national security.
This directive emerged during a phase of amplified governmental scrutiny concerning advanced AI technologies, largely due to apprehensions that such systems could potentially be weaponized by military intelligence in adversarial nations like China and Russia.
Faced with the challenge of distinguishing between domestic and foreign users in real time, Anthropic decided to suspend access for all users, including Americans.
Introduced in early June, Fable 5 represents a curated public version of Mythos 5, a model that Anthropic had previously withheld from broader public use due to its exceptional capability to detect vulnerabilities that could be exploited by cybercriminals.
Mythos 5 was made available only to a select group of businesses and governmental entities as part of what Anthropic refers to as Project Glasswing, aimed at allowing these exclusive stakeholders to identify other possible hazardous applications or weaknesses in their own systems.
On June 26, Anthropic reinstated limited access to Mythos 5 for a small set of ‘trusted’ U.S. organizations. Subsequently, on June 30, the Commerce Department completely lifted restrictions on both models.
Repercussions in the AI Landscape
The restrictions stirred ripples throughout the AI sector, igniting debates about the applicability of export controls in regulating AI model accessibility.
According to Reuters, the order from the U.S. government followed disclosures by Amazon researchers, who discovered a means to circumvent Fable 5’s safeguards, thereby enabling the powerful AI to pinpoint software vulnerabilities or illicit information.
In response, Anthropic has instituted a novel safeguard designed to block this behaviour, routing any banned requests to its earlier Opus 4.8 model.
While this decision may displease some users, Anthropic expressed that it was necessary to preserve the broader functionality of Fable 5.
The company cautioned, however, that making any AI model fully impervious to “jailbreaks”—techniques that bypass or disable safeguards—is likely an insurmountable challenge.
“Minor jailbreaks will arise, some potentially harmful, and while no comprehensive jailbreaks for Fable 5 have been identified at present, specialized safety researchers are continually assessing it,” stated the company.
Increased Cooperation with Government
Further reinforcing its position, Anthropic declared its intention to deepen collaboration with the U.S. government by granting designated government partners extended early access to its models.
A correspondence from U.S. Commerce Secretary Howard Lutnick indicated that Anthropic has committed to working closely with the government on safety protocols for Mythos, Fable, and future offerings, while also pledging to notify authorities of any malevolent activities.
Notably, Lutnick reserved the right to revisit these decisions and possibly reimpose licensing requirements should circumstances dictate or if Anthropic fails to uphold its commitments.
The initiation of intensified scrutiny on AI models coincided with an executive order signed by former President Trump in June, which established a framework for AI developers to present “covered frontier models” to the government for a limited period before unveiling them to trusted partners.
Competing AI firm OpenAI similarly faced restrictions, delaying the public launch of GPT-5.6 at the request of the U.S. government and limiting its access to a vetted core of partners.
This episode signifies a shift in Washington’s approach to AI policy. The invocation of export controls, traditionally applied to hardware underpinning AI models, now finding application to software, indicates a burgeoning trend in governmental intervention within the AI realm.
The Context of Concerns
Isaac Harris, executive director for the Frontier Security Institute, a nonprofit devoted to AI and national security, suggested that there appears to be a developing framework for standards surrounding U.S. AI models.
Nonetheless, he noted that significant ambiguity remains regarding how significantly risky capabilities emerging from China, which feature fewer regulatory safeguards, will be managed in the U.S. market.
Ongoing Tensions with the U.S. Government
This development is but a recent chapter in a lengthy and contentious saga between Anthropic and the U.S. government.
Dean Ball, a former AI advisor encompassing the Trump administration’s 2025 AI Action Plan, recounted that the collaboration between Anthropic and the Pentagon evolved after agreeing to use Claude in classified operations, including intelligence analysis—albeit with stipulations against domestic mass surveillance and fully autonomous weaponry.
However, as relationships soured, tensions escalated when Emil Michael, the under secretary of war for research and engineering, requested the removal of clauses prohibiting the use of Claude for analyzing vast swathes of commercially collected data—a core issue tied to mass surveillance qualms.
As the clash became increasingly personal and political, U.S. Defense Secretary Pete Hegseth publicly criticized Anthropic, labeling its motives as veering towards the manipulation of military operational decisions, which he deemed unacceptable.
This conflict reached a zenith in February when Hegseth threatened to deploy the Defense Production Act to officially designate Anthropic as a “supply chain risk.”
This designation is typically reserved for entities deemed too hazardous for the U.S. military’s supply chain, historically reserved for foreign firms like Huawei, due to espionage fears.
Trump subsequently declared a cessation of all federal engagement with Anthropic technologies, warning of significant repercussions should the company fail to cooperate during a transition away from its systems.
Despite these challenges, Anthropic remains resolute, asserting that such coercive actions are unfounded and will not deter its stance on mass surveillance or automatic weaponry.
In what seems a striking shift, the Pentagon has since signed a contract with OpenAI, which affirms similar ethical boundaries against military applications deemed unacceptable by Anthropic.
Amid these developments, Anthropic has initiated federal lawsuits, challenging the designation of “supply chain risk” as a retaliatory measure against its advocacy for AI safety.
As this legal battle unfolds, the broader implications for AI policy and its intersection with national security remain precarious and multifaceted.
Source link: Hindustantimes.com.
