The White House is demanding that Anthropic actively block every possible jailbreak in its most advanced AI models, a directive that may be technically impossible to fulfil.
Administration officials have told sources that if Anthropic intends to relaunch Claude 5—the model recently taken offline due to export controls over fears of prompt-based bypasses—it must first demonstrate a concrete solution to the alleged security vulnerabilities. While Anthropic has repeatedly argued that these threats are exaggerated and that the practical impact of such bypasses is negligible, it reinforced this stance during a technical briefing with the Commerce Department and Office of the National Cyber Director, Sean Cairncross, on Monday.
However, government officials claim they have moved past debating the severity of the issue. The National Security Agency has already identified methods to disable the guardrails on Claude 5, which are designed to prevent users from accessing capabilities related to cybersecurity, chemistry, and biology within the underlying Mythos model.
According to three sources familiar with the discussions, the administration now views this as a problem for Anthropic to resolve alone. Neither the Commerce Department’s Center for AI Standards and Innovation nor the National Security Agency possesses the personnel capacity to hunt down every conceivable jailbreak across all models entering the market.
Consequently, officials believe Anthropic must take a more proactive role, continuously testing not only Claude 5 but its entire frontier model suite to identify potential bypasses and reporting them directly to the government.
Yet, a fundamental question remains unanswered: how exactly is Anthropic supposed to prevent jailbreaking? Independent cybersecurity experts increasingly argue that model guardrails are merely a temporary fix, as skilled users and future AI systems will inevitably find ways to circumvent constraints. This reality suggests the administration’s desired outcome cannot be achieved.
A White House spokesperson declined to comment on the matter.
DNI = Do Not Invite
Bill Pulte, Trump’s initial choice for Acting Director of National Intelligence, was on track to never assume the role before the administration pivoted. Now, the permanent nominee, Jay Clayton, faces the prospect of never serving in the position.
Initially, Trump selected Pulte, his housing finance chief, to replace outgoing DNI Tulsi Gabbard. However, bipartisan opposition arose because Pulte lacked the statutory national security experience required for the role and had flagged questionable mortgage fraud accusations against Trump’s political rivals. In response, Trump announced Clayton, the US attorney for the Southern District of New York, as his permanent nominee.
Gabbard was scheduled to leave on June 18, with Pulte set to start on June 19. Senate Republicans had considered whether fast-tracking Clayton’s hearing to June 17 and allowing him to start by June 22 would leave Pulte without a seat in the building.
On Wednesday, Trump dismantled the plan. Amidst a broader feud with Senate Republican leadership regarding the filibuster, he announced Clayton’s hearing would be delayed indefinitely, seemingly to prevent Pulte from being sidelined. Senate Republicans subsequently stated the hearing would proceed unless Clayton failed to appear or withdrew his nomination.
This situation could prove damaging for the Office of the Director of National Intelligence, which Trump has directed Pulte to significantly downsize. Staffers have expressed dissatisfaction with what they perceive as Pulte’s minimal effort to acquaint himself with the agency and his lack of regular briefings.
Internally, critics have suggested Pulte appeared eager to secure the perks of the role, such as a security detail and government jet travel, without undertaking the often thankless work of producing sober intelligence briefings and coordinating the nation’s intelligence agencies.
A White House spokesperson declined to comment, directing inquiries to Trump’s announcement on Truth Social.
Influence at the UFC
Last week, sources indicated that White House aides anticipated a gathering of donors and corporate executives at UFC Freedom 250 over the weekend, aiming to facilitate proximity to Trump and senior administration officials.
The donors and executives duly attended, both on fight night and at various parties held on the sidelines.
Notable attendees included Paramount CEO David Ellison, who received approval from the Justice Department’s antitrust division last week to acquire Warner Bros Discovery. Also present was Meta CEO Mark Zuckerberg, who was observed speaking with Trump at one point.
Earlier in the weekend, Meta hosted a private party at the Ned’s Club overlooking the White House. Guests included several Trump officials: acting attorney general Todd Blanche, White House deputy chief of staff James Blair, White House press secretary Karoline Leavitt, and interior secretary Doug Burgum.
Other attendees included members of the Trump family—Jared Kushner, Ivanka Trump, and Kai Trump—who mingled with guests such as former Clinton strategist Adrienne Elrod, Axios cofounders Mike Allen and Jim VandeHei, Fox News host Shannon Bream, Washington AI network founder Tammy Haddad, and former Trump strategist Kellyanne Conway.
Key takeaways
The White House is insisting Anthropic proactively block all jailbreaks in its frontier models, a task independent experts argue is technically unfeasible given the nature of AI constraints.
Bill Pulte, Trump’s initial pick for Acting DNI, faces an indefinite delay in assuming the role, while permanent nominee Jay Clayton’s confirmation hearing has been stalled indefinitely.
Corporate leaders including David Ellison and Mark Zuckerberg attended UFC Freedom 250, using the event to network with Trump administration officials and family members.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
