The trigger of the incident starts with Anthropic
Previously, negotiations between Anthropic and the Pentagon had reached an impasse. The Department of Defense required AI companies to accept an "any lawful use" clause, allowing models to be used freely under U.S. law, including potential domestic surveillance and autonomous weapons systems. Anthropic CEO Dario Amodei explicitly stated that the company "could not, in good conscience," accept this requirement, insisting on two core red lines:
- Prohibition of domestic mass surveillance of U.S. citizens
- Ensuring human ultimate responsibility for the use of lethal force (including autonomous weapons systems)
As a result, Anthropic was officially designated as a "supply chain risk" by the Pentagon, facing usage restrictions in federal systems and with contractors, and potentially being gradually excluded from government-related business. Anthropic subsequently filed a lawsuit, claiming the move was unlawful.
Just hours after Anthropic faced pressure from the blacklist, OpenAI quickly announced its own agreement. Altman emphasized that OpenAI's agreement also included safety guardrails:
Our most important safety principles include prohibiting domestic mass surveillance and ensuring human responsibility for the use of force (including autonomous weapons systems). The Pentagon agreed to these principles and has incorporated them into the agreement.
However, this statement did not quell the controversy; instead, it intensified internal divisions.
The reaction from OpenAI employees was particularly intense
Dozens of current employees, together with colleagues from Google, signed an open letter titled "We Will Not Be Divided," calling on company leadership to stand with Anthropic and reject the Pentagon's current demands. The letter explicitly stated:
Models must not be allowed for 'domestic mass surveillance' and 'autonomous killing without human oversight.'
The open letter quickly gathered hundreds of signatures, reflecting deep concerns among employees about AI military applications. History seems to be repeating itself — eight years ago, Google faced massive employee protests over the Project Maven military project, eventually partially withdrawing; now, a similar ethical storm is hitting OpenAI.
The user-side reaction was more direct and quantifiable
According to data from market intelligence firm Sensor Tower, on February 28 (the day after the contract announcement), the single-day uninstall rate of the U.S. ChatGPT mobile app surged 295% compared to the previous day, far exceeding the average fluctuation of about 9% over the past 30 days. At the same time, downloads dropped significantly — 13% on Saturday and another 5% on Sunday. A large number of users publicly stated they were "switching to Claude," and downloads of Anthropic's Claude app rose sharply, momentarily surpassing ChatGPT on Apple's App Store free app charts. On social media, hashtags like "#CancelChatGPT" and "Switch to Claude" quickly gained traction. Protest graffiti even appeared outside the OpenAI office, while messages outside the Anthropic office were mostly supportive.
Facing pressure, OpenAI did not fully retreat but made adjustments. A few days later, the company announced modifications to the agreement, adding explicit clauses: AI systems "shall not be intentionally used for domestic surveillance of U.S. personnel or nationals," citing relevant federal laws. Altman acknowledged in internal memos and an all-hands meeting that the company
had intended to help ease the situation and avoid worse outcomes, but it appeared opportunistic and hasty.
He also stated that the company "cannot make operational decisions about how the Pentagon uses the technology," citing examples like a strike on Iran or an invasion of Venezuela to argue that companies should not intervene in specific government military actions. This statement further fueled the controversy — employees and users questioned: if the company cannot control the end use, why sign the agreement?
At the core of this incident is the profound conflict between ethics and business faced by AI companies.
From an ethical perspective
The militarization of AI technology does pose real risks. If powerful models are used for domestic mass surveillance, they could violate citizens' privacy. If used for autonomous weapons systems (lethal autonomous weapons), they could lead to uncontrollable lethal decisions, violating the international humanitarian principle that "humans must always maintain ultimate control." Many employees and users believe that AI companies, as technology creators, have a responsibility to set hard guardrails, rather than simply relying on "lawful use" clauses. Anthropic's refusal is a steadfast adherence to this responsibility, earning it praise as a "principled company."
From a business perspective
Cooperation with the government, especially the Department of Defense, means huge contracts, data resources, and strategic influence. Refusing could lead to business losses, supply chain exclusion, and even being labeled "uncooperative" in the national security narrative. OpenAI's choice to negotiate and partially compromise reflects practical considerations. However, this "opportunistic" approach has cost the company user trust. Data shows that ChatGPT has seen significant user churn in the short term, while Claude has risen on the momentum, subtly shifting the competitive landscape.
The focus of the debate is highly concentrated on "military use restrictions." Supporters argue that AI should serve national security and operate within the legal framework; opponents emphasize the need for clear, enforceable red lines to prevent technology from being abused or spiraling out of control. Similar controversies have previously arisen at companies like Google and Microsoft, and this concentrated eruption between OpenAI and Anthropic reflects that the entire frontier AI industry is at a crossroads.
For the global AI ecosystem
This event serves as an important wake-up call. It reminds companies: users and employees are increasingly concerned about technology ethics, and purely pursuing business or government contracts can trigger a crisis of trust. Anthropic's persistence, though costly in the short term, may earn it reputation and user loyalty in the long run. OpenAI's quick adjustments show that companies can still mitigate pressure through transparent communication and agreement optimization, but fundamentally, more mature internal governance mechanisms are needed — such as independent ethics committees and public impact assessments.
From a broader perspective
The military application of AI is a microcosm of great power technology competition. The U.S. is pushing frontier models into the defense system to maintain its edge; China and other countries are also accelerating their deployments. Finding a balance between security and ethics will become a global challenge. Over-commercialization may lead to the erosion of ethical bottom lines, while excessive restrictions may hinder technological progress and national security. The ideal path is to establish a transparent, multi-stakeholder framework: companies set core red lines, governments provide clear legal boundaries, and the public and employees participate in oversight.
Looking ahead
Similar conflicts may occur frequently. More AI companies (including Google, xAI, etc.) have already faced or will face similar cooperation opportunities with the Pentagon. User choice, employee voices, and corporate social responsibility will jointly shape the industry's direction. The showdown between OpenAI and Anthropic is not a simple win-or-lose but a touchstone of industry maturity.
In summary, the OpenAI Pentagon contract incident has pulled AI ethics from abstract discussion into a real-world arena. The employee backlash, user exodus, and praise for Anthropic all remind us: no matter how powerful technology becomes, it must serve human well-being, not turn against its creators. The future of AI depends on how we define and safeguard those non-negotiable red lines today.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接