Key developments
Pentagon contract shifts after Anthropic refusal
Yahoo Tech's Becca Caddy reports that the Pentagon asked Anthropic to remove contract guardrails banning domestic mass surveillance and fully autonomous weapons. Anthropic declined, and the contract reportedly went to OpenAI instead. The report turns a procurement dispute into a concrete test of how far defense buyers will push AI vendors on privacy and autonomy limits.
Why it matters
It shows surveillance and weaponization limits are now being negotiated directly in government AI contracting.
Sources & driving stories
YAHOO TECH · Becca Caddy
Yahoo Tech coverageCOTI Nightfall advances toward mainnet
AInvest's Penny McCormer says Vitalik Buterin is backing privacy-preserving infrastructure built around garbled circuits and zero-knowledge proofs. The piece says COTI Nightfall is planned to start with a testnet soon, then move to mainnet in 2026, with use cases including private transfers, tokenized real-world assets, regulated payments and private DeFi. The article argues the project will need measurable Ethereum transaction volume to prove commercial durability.
Why it matters
It gives a concrete deployment timeline for one of the more visible privacy layers being positioned for Ethereum.
Sources & driving stories
AINVEST · Penny McCormer
AInvest coverageWorth noting
WORTH NOTING
15% of AI agent skills may be malicious
The cited estimate quantifies how hidden instructions could compromise AI assistants and leak data in privacy-sensitive workflows.
Still unclear
OPEN QUESTION
Will defense contracts now require explicit surveillance and autonomy carveouts?
The Anthropic episode suggests procurement terms may become the main mechanism for limiting privacy-invasive AI uses.
OPEN QUESTION
Can privacy rollups generate enough real transaction volume?
Nightfall's credibility depends on moving beyond a roadmap to sustained on-chain usage.