The U.S. government opposes Anthropic's efforts to expand access to its AI model, Mythos. This model is capable of identifying and exploiting software vulnerabilities, and its widespread adoption could potentially be leveraged by malicious actors for cyberattacks. The Trump administration has explicitly opposed the move, citing security concerns, while some officials also worry that insufficient computing power might impact governmental usage. The two sides have previously clashed over AI tools for the Pentagon, with political divisions and personnel disputes further escalating tensions.
The tensions between the White House and artificial intelligence company Anthropic have escalated once again. Amid ongoing cybersecurity concerns, the Trump administration has explicitly opposed Anthropic's efforts to expand the use of its AI model Mythos. This stance not only hinders the commercial rollout of the model but also reflects the unresolved structural conflicts between the government and Silicon Valley AI firms.
According to The Wall Street Journal citing insiders, Anthropic recently proposed granting access to Mythos to approximately 70 additional enterprises and institutions, which would bring the total number of entities authorized to use the model to around 120. Officials from the Trump administration explicitly opposed this move, citing security concerns. Some White House officials also worry that Anthropic may lack sufficient computing power to serve such a large user base simultaneously, potentially affecting the government’s ability to effectively utilize the model.
This confrontation occurs against a backdrop of already heightened tensions. Anthropic and the White House previously engaged in a public dispute over the use of AI tools by the Pentagon, with related litigation progressing in two separate courts. Meanwhile, Anthropic announced last week that it is investigating an alleged unauthorized access incident involving Mythos, further exacerbating external concerns about the proliferation of large-scale software vulnerabilities and significantly raising market awareness of AI-related security risks.
Mythos Security Concerns: Exploitation of Vulnerabilities Raises Alarms
The core reason for the White House's heightened focus on the Mythos model lies in its prominent capabilities in the field of cyber offense and defense. Cybersecurity experts point out that the model can identify and exploit software vulnerabilities, possessing the potential to launch cyberattacks and cause widespread disruption online—a feature that has deeply unsettled multiple government agencies and corporations.
In early April this year, Anthropic opened the preview version of Mythos to select top-tier enterprises and institutions, sparking global concerns and prompting various parties to begin assessing and mitigating the potential security risks associated with a broader rollout of the model. Currently, initial access permissions cover approximately 50 enterprises and institutions responsible for managing critical infrastructure. While the Trump administration is working to expand access for government agencies, it does not support broader commercial promotion.
Several leading AI companies, including OpenAI and Google under Alphabet, have provided early access to their models for security researchers to discover and patch vulnerabilities before they can be maliciously exploited. However, according to The Wall Street Journal, industry insiders are deeply concerned about the sheer volume of vulnerabilities being discovered, along with the subsequent workload involved in creating, testing, and installing patches.
Deep Rift: Political Divisions and Personnel Controversies
The conflict between the White House and Anthropic is not limited to security issues; political differences also constitute a significant friction point. The Trump administration has criticized Anthropic both for its ties to liberal funders who advocate strict AI regulation and for employing several former officials from the Biden administration.
Personnel-related frictions have also been specific and evident. According to reports citing insiders, Anthropic’s former researcher Collin Burns was initially expected to assume a government position specifically tasked with evaluating top-tier AI models. However, the government abruptly changed plans and sought someone else. Insiders stated that senior White House officials were not informed beforehand about the hiring arrangement for Burns and were unwilling to see individuals from leading AI companies take on roles requiring close collaboration with major enterprises.
Computational Power Bottleneck: A Realistic Constraint on Business Expansion
In addition to political maneuvering, insufficient computational power also serves as an objective obstacle limiting Mythos’ expansion. Some White House AI advisors, including venture capitalist David Sacks, have speculated that Anthropic’s decision to pursue a limited rollout is partly due to its computational resources falling short of competitors’. Computational power is the core resource enabling AI model users to execute various tasks.
To address this shortfall, Anthropic has recently entered into agreements with$Amazon (AMZN.US)$, Google, and$Broadcom (AVGO.US)$to expand its computational capacity. However, these projects will require some time before they are officially operational. AI companies can typically prioritize the allocation of computational resources to specific user groups such as government entities. The industry generally anticipates the emergence of more efficient models to fundamentally alleviate computational power shortages.
Calls for Collaboration Amidst Divisions: Practical Pressures Beneath Disagreements
Despite ongoing tensions, stakeholders continue to advocate for collaboration. Dean Ball, who served as an AI advisor during the Trump administration and is now a senior fellow at the American Innovation Foundation, pointed out that Anthropic's opposition to the Trump administration is counterproductive. Both sides have a critical need to cooperate on core issues like cybersecurity, and adversarial posturing would only yield negative outcomes.
White House officials stated that the government is balancing innovation with security while actively collaborating with the private sector to ensure the safe deployment of AI models. This statement leaves room for repairing relations between the two parties. However, for now, the White House’s rejection of Anthropic’s request to expand access to Mythos clearly indicates that their contest continues, with significant disagreements far from resolved.
Editor/Rocky