Anthropic, a Threat to the Pentagon: Tensions Surrounding AI

⚡

Key Takeaways

1The Pentagon has classified Anthropic as a supply chain risk, halting the use of Claude in its projects.

2OpenAI has signed an agreement with the DoD, triggering a "cancel ChatGPT" trend and a boycott with a 295% increase in uninstalls.

3OpenAI has launched GPT-5.4, improving token efficiency and reducing errors by 33% compared to GPT-5.2.

💡Why it matters — These decisions highlight the growing tensions between technological innovation and national security concerns.

The Pentagon Designates Anthropic as a Risk

The Pentagon has recently designated Anthropic as a potential risk to its supply chain. This decision requires defense suppliers and contractors to certify that they do not use Anthropic's models, including Claude, in their work for the Department of Defense (DoD). This measure has led to an immediate cessation of the use of Anthropic technologies by the Pentagon.

The controversy arose from the terms of the contract between Anthropic and the DoD. Anthropic insisted on establishing clear red lines against domestic mass surveillance and the use of fully autonomous weapons, while the DoD requested access for all legal uses. Anthropic plans to contest this designation in court, arguing that it should only apply to uses directly related to DoD contracts and that it violates the requirement for the least restrictive means.

Reactions and Clarifications

Following this designation, Defense Secretary Pete Hegseth suggested that no military partner should engage in commercial activities with Anthropic, which caused confusion regarding the extent of this ban. In response, cloud providers like Microsoft, Google, and AWS clarified that Claude remains accessible for non-defense-related workloads through their respective platforms such as M365, GitHub, AI Foundry, and Google Cloud.

Meanwhile, OpenAI announced an agreement with the DoD to deploy its models in classified environments. This agreement explicitly prohibits mass surveillance and the use of autonomous weapon systems.

The Rise of "Cancel ChatGPT"

The signing of the agreement between OpenAI and the DoD has sparked a wave of criticism, fueling the "cancel ChatGPT" trend. OpenAI had to adjust its agreement to clarify that its AI system should not be used for domestic surveillance of American citizens. This discontent has led to a boycott titled "QuitGPT," with protests outside OpenAI's headquarters in San Francisco. This mobilization resulted in a 295% increase in ChatGPT uninstalls.

Launch of GPT-5.4 by OpenAI

OpenAI has recently launched GPT-5.4, offered in three variants: standard, Pro, and Thinking. This model is presented as more powerful and efficient for professional tasks. The GPT-5.4 API now supports a context of up to 1 million tokens, and token efficiency has been improved, allowing the same tasks to be accomplished with fewer tokens than previous versions. GPT-5.4 has also achieved record results on computing benchmarks, reducing errors by 33% compared to GPT-5.2.

Other Developments in the AI Sector

Luma has launched creative AI agents based on its new "Unified Intelligence" models, capable of generating and refining multimedia content. Anthropic reported a widespread outage affecting access to Claude.ai and Claude Code, although the API remained functional.

On its part, Google has made its Gmail, Drive, and Docs services "agent-ready" with a new CLI tool to facilitate the integration of AI agents. Alibaba is facing internal discussions following the sudden departure of its Chief Technology Officer, Qwen.

Finally, OpenAI raised $110 billion in a historic private funding round, and Cursor surpassed $2 billion in annualized revenue, with 60% of that revenue coming from enterprise clients.

Advances in Research

A new video reasoning dataset, the VBVR-Dataset, has been created. It includes 1,007,500 clips and covers 200 tasks across five cognitive pillars.