Anthropic Resumes Access to AI Model Following Government Concerns
In a significant development, Anthropic has reinstated customer access to its advanced AI model, Fable, after enduring a hiatus of over two weeks.
This disruption was instigated by apprehensions from the US government regarding the potential misuse of the technology, particularly in relation to severe cyber threats.
The San Francisco-based firm announced that export restrictions have been lifted, referencing a statement made by the US Commerce Secretary, Howard Lutnick.
He remarked, “During the past fortnight, we collaborated closely with Anthropic to scrutinize and authorize Fable 5, ensuring coherence with US Government standards and fortifying America’s preeminence in AI.”
Just three days post its launch on June 9, the US administration mandated Anthropic to curtail access to Fable 5 and a companion model, Claude Mythos 5.
This directive prohibited their use by foreign nationals, both domestically and abroad. The rationale behind such stringent measures stemmed from concerns that these powerful AI systems might facilitate catastrophic cyberattacks orchestrated by foreign entities.
On Friday, the government softened its stance, permitting Anthropic to distribute Mythos 5 to select “trusted” US organizations, partially rescinding the earlier order.
This release was confined to a limited number of vetted partners, specifically for defensive cybersecurity applications.
Lutnick, in a correspondence to Anthropic that was reviewed by Reuters, emphasized, “Anthropic has acquiesced to proactively identify and mitigate security risks linked to the models, collaborate vigilantly with the US government on protocols and standards for Mythos, Fable, and forthcoming models, and report any malicious activities to the authorities.”
The imposition of the ban was perceived as an escalation in the Biden administration’s approach towards the oversight of domestic AI enterprises.
Previously, a more lenient framework had prevailed, motivated by a competitive drive to outpace China in the quest for AI supremacy.
The government’s decisive action followed revelations in a research report by Amazon, which disclosed a method to circumvent Fable 5’s safeguards, enabling the identification of software vulnerabilities. This finding illuminated the potential for exploitation by a nefarious actor.
In response to the heightened export controls, Anthropic initially withdrew its AI offerings. However, they have since augmented their security protocols, asserting in a blog post, “Researchers from the US Department of Commerce’s Center for AI Standards and Innovation (CAISI) have evaluated both our former and newly implemented safeguards, confirming their robustness.”
On Tuesday, Anthropic announced plans to “enhance our collaboration with government entities” and establish dedicated teams to address shared priorities with governmental stakeholders.
Recently, intelligence agencies in Australia, the US, the UK, New Zealand, and Canada issued a stark warning. While acknowledging the potential of AI to bolster cyber defenses, they cautioned that it also expedites the emergence of sophisticated cyber threats.
The Five Eyes agencies articulated, “Next-generation AI models are projected to surpass current industry standards, fundamentally revolutionizing both offensive and defensive cyber capabilities. The timeline for this advancement is imminent; we project it to occur within months, rather than years.”
Susie Wiles, Chief of Staff to Donald Trump, expressed her appreciation for companies collaborating with the White House to execute the President’s Executive Order on “Promoting Advanced AI Innovation and Security.”
She noted that these efforts encompass exemplary actions pertaining to advanced model access and security testing.
Wiles added, “The collaboration between government and private sectors has reached unprecedented levels, reinforcing an America First ethos. Our collective priority remains clear: to deploy the most advanced technology expeditiously and safely.
However, the US government’s procedures for vetting companies eligible for access to these AI models have faced criticism.
OpenAI’s CEO, Sam Altman, articulated his reservations, stating, “While extensive safety testing is prudent, I oppose the notion of the government determining which customers gain access.”
Following government requests, the launch of GPT-5.6 has been delayed, limiting its availability to a select group of pre-approved partners.
Source link: Theguardian.com.
