A simmering disagreement between the Trump administration and AI developer Anthropic over its most advanced AI models appears to be nearing a critical juncture. At the heart of the dispute is Claude Fable 5, an AI model that the administration pulled offline last week using export controls.
Officials assert that if Anthropic wishes to re-release Fable 5, the company must proactively address alleged vulnerabilities related to “jailbreaking.” This technique involves using specific prompts to bypass an AI model’s built-in safeguards, a concern the government takes very seriously.
Anthropic’s AI Jailbreak Challenge
Anthropic has consistently maintained that the administration’s concerns about Fable 5’s jailbreaking susceptibility are exaggerated and that their impact is minimal. The company reiterated this stance during a technical meeting with the Commerce Department and the Office of the National Cyber Director, Sean Cairncross, earlier this week.
However, government officials suggest they are beyond debating the significance of these jailbreaks. The National Security Agency (NSA) has concluded that methods exist to disable Fable 5’s guardrails, which are designed to prevent users from accessing powerful capabilities of its underlying Mythos model related to cybersecurity, chemistry, and biology.
The administration views this as Anthropic’s responsibility to resolve. Sources familiar with the discussions indicate that neither the Commerce Department’s Center for AI Standards and Innovation nor the NSA possesses the resources to constantly pursue every potential jailbreak across all new AI models.
Consequently, the government expects Anthropic to adopt a more proactive approach. This involves continually testing not just Fable 5 but all of its cutting-edge AI models for potential jailbreaks and transparently reporting these findings to the authorities.
Despite this expectation, a fundamental question remains: how exactly is Anthropic supposed to prevent jailbreaking? Many independent cybersecurity experts increasingly view AI model guardrails as a temporary measure.
They argue that skilled users, and even future AI models, will inevitably discover ways to bypass these constraints. This suggests that the White House’s current demands might be technically unfeasible in the long term, posing a significant challenge for AI developers.
DNI Shake-Up: Pulte vs. Clayton
A political drama is unfolding within the Trump administration regarding the appointment of the next Director of National Intelligence (DNI). What initially seemed like a clear path for Trump’s pick, Bill Pulte, as Acting DNI, has become a tangled situation, potentially sidelining the permanent nominee, Jay Clayton.
Originally, Trump designated Pulte, his housing finance chief, to succeed outgoing DNI Tulsi Gabbard. This decision faced bipartisan opposition due to Pulte’s lack of national security experience, a statutory requirement for the role, and allegations of questionable mortgage fraud accusations he flagged against Trump’s political adversaries.
In response to the backlash, Trump then announced Jay Clayton, the U.S. Attorney for the Southern District of New York, as his nominee for the permanent DNI position. With Gabbard scheduled to depart on June 18th and Pulte set to begin on June 19th, Senate Republicans pondered whether Clayton could be fast-tracked to start by June 22nd, effectively preventing Pulte from ever assuming the role.
However, Trump abruptly altered this plan mid-week. Amidst a broader dispute with Senate Republican leadership over the filibuster, Trump indefinitely delayed Clayton’s hearing. This move appears to be a deliberate effort to ensure Pulte’s appointment.
Senate Republicans quickly countered, announcing that the hearing would proceed unless Clayton failed to appear or his nomination was withdrawn. This ongoing battle creates significant uncertainty for the Office of the Director of National Intelligence (ODNI).
Staffers at the ODNI are reportedly dismayed by what they perceive as Pulte’s minimal engagement with the agency and his lack of regular briefings. They have privately expressed concerns that Pulte seems more interested in the perks of the DNI role—such as a security detail and government jet travel—than in the demanding work of providing intelligence briefings and coordinating national intelligence agencies. A White House spokesperson declined to comment, referencing Trump’s announcement on Truth Social.
High-Profile Attendees at UFC Freedom 250
As previously anticipated, last weekend’s UFC Freedom 250 event served as a magnet for donors and corporate executives eager to connect with Trump and senior administration officials. These influential figures were indeed present, not just on fight night but also at a series of exclusive parties held in conjunction with the event.
Among the notable attendees was Paramount CEO David Ellison, who recently received antitrust approval from the Justice Department to acquire Warner Bros. Discovery. Also spotted was Meta CEO Mark Zuckerberg, who was seen engaging in conversation with Trump himself.
Earlier in the weekend, Meta hosted a private gathering at the Ned’s Club, offering views of the White House. This exclusive event attracted several Trump officials, including acting attorney general Todd Blanche, White House deputy chief of staff James Blair, White House press secretary Karoline Leavitt, and interior secretary Doug Burgum.
Trump family members, including Jared Kushner, Ivanka Trump, and Kai Trump, also mingled with guests. The eclectic guest list further included former Clinton strategist Adrienne Elrod, Axios co-founders Mike Allen and Jim VandeHei, Fox News host Shannon Bream, Washington AI network founder Tammy Haddad, and former Trump strategist Kellyanne Conway.
Source: Wired – AI
