# AI News Digest - 2026-03-02

1. The Pentagon, OpenAI, and Anthropic disclosed contract details and faced public fallout over the contract provision described as "all lawful use."

2. ETH Zurich and Anthropic researchers demonstrated that commercially available AI models could link pseudonymous online names to real identities in minutes for a few dollars per person.

3. Frontier LLMs including GPT-5 and Claude 4.6 were reported to lose up to 33% accuracy as conversation length increased.

4. ElevenLabs and Google dominated Artificial Analysis' updated speech-to-text benchmark, ranking highest in the evaluation.

5. Moltbook was found to host over 2.6 million AI agents that interacted without human involvement and showed no evidence of mutual learning, shared memory, or social structures.

# References

1. [https://the-decoder.com/the-pentagon-openai-anthropic-fallout-comes-down-to-three-words-all-lawful-use/](https://the-decoder.com/the-pentagon-openai-anthropic-fallout-comes-down-to-three-words-all-lawful-use/)

[The Pentagon-OpenAI-Anthropic fallout comes down to three words: "all lawful use"](https://the-decoder.com/the-pentagon-openai-anthropic-fallout-comes-down-to-three-words-all-lawful-use/)

1. [https://the-decoder.com/ai-can-link-fake-online-names-to-real-identities-in-minutes-for-just-a-few-dollars/](https://the-decoder.com/ai-can-link-fake-online-names-to-real-identities-in-minutes-for-just-a-few-dollars/)

[AI can link fake online names to real identities in minutes for just a few dollars](https://the-decoder.com/ai-can-link-fake-online-names-to-real-identities-in-minutes-for-just-a-few-dollars/)

1. [https://the-decoder.com/even-frontier-llms-from-gpt-5-onward-lose-up-to-33-accuracy-when-you-chat-too-long/](https://the-decoder.com/even-frontier-llms-from-gpt-5-onward-lose-up-to-33-accuracy-when-you-chat-too-long/)

[Even frontier LLMs from GPT-5 onward lose up to 33% accuracy when you chat too long](https://the-decoder.com/even-frontier-llms-from-gpt-5-onward-lose-up-to-33-accuracy-when-you-chat-too-long/)

1. [https://the-decoder.com/elevenlabs-and-google-dominate-artificial-analysis-updated-speech-to-text-benchmark/](https://the-decoder.com/elevenlabs-and-google-dominate-artificial-analysis-updated-speech-to-text-benchmark/)

[ElevenLabs and Google dominate Artificial Analysis' updated speech-to-text benchmark](https://the-decoder.com/elevenlabs-and-google-dominate-artificial-analysis-updated-speech-to-text-benchmark/)

1. [https://the-decoder.com/moltbooks-alleged-ai-civilization-is-just-a-massive-void-of-bloated-bot-traffic/](https://the-decoder.com/moltbooks-alleged-ai-civilization-is-just-a-massive-void-of-bloated-bot-traffic/)

[Moltbook's alleged AI civilization is just a massive void of bloated bot traffic](https://the-decoder.com/moltbooks-alleged-ai-civilization-is-just-a-massive-void-of-bloated-bot-traffic/)

For the site tree, see the [root Markdown](https://ixtj.dev/.md).