01 Meta pushes deeper into AI
Hacker News AI highlighted Meta will train AI agents by tracking employees' mouse, keyboard use as a main item.
Related coverage also includes Meta will train AI agents by tracking employees' mouse, keyboard use, making this look more like a broader theme than a stray post.
The cluster is corroborated across 2 sources, so it reads as a developing trend rather than a one-off mention.
- Meta will record employee mouse and keyboard activity to create interactive training data for AI agents.
- Interactive traces from internal work are valuable because public datasets often lack step‑by‑step behavior examples.
- Collecting this data brings privacy and governance tradeoffs that companies must manage before using it to train models.
02 OpenAI launches Codex Labs and partners to scale coding AI for enterprises
OpenAI announced a new Codex Labs initiative and named consulting partners — including Accenture, PwC, and Infosys — to help enterprises deploy and scale Codex across the software development lifecycle. The company also reported Codex has reached 4 million weekly active users (WAU).
Codex Labs appears aimed at standardizing enterprise onboarding, integration, and best practices so organizations can apply code generation across testing, refactoring, and developer tooling. Partnering with established systems integrators signals OpenAI’s push to move Codex from pilot projects into broader corporate infrastructure.
For businesses this means more formal support and pathways to adopt code‑generation tools, but it also raises standard questions about guardrails, IP handling, and how generated code will be validated in production environments.
- OpenAI introduced Codex Labs and tapped Accenture, PwC, and Infosys to help enterprises deploy Codex.
- Codex reached 4 million weekly active users, underscoring adoption among developers.
- Enterprise rollouts will need practices for validation, IP management, and integration with existing development workflows.
03 Big defense spending and cyber concerns: Pentagon drone ask and Anthropic’s Mythos in government use
Ars Technica reports the Pentagon has requested roughly $54 billion for drone investments — an amount that rivals the military budgets of many nations and, the article notes, is comparable to Ukraine’s entire military budget. The proposal underscores how acquisition priorities are shifting toward unmanned systems at scale.
On the model front, reporting shows U.S. agencies have been engaging with commercial AI in uneven ways: TechCrunch says the NSA is reportedly using Anthropic’s restricted Mythos model, even as public disputes have flared between Anthropic and parts of the defense establishment. Meanwhile, Ars Technica flagged concerns that Mythos could accelerate cyber‑attack discovery faster than defenses can adapt.
Taken together, the stories illustrate two trends: governments are budgeting at scale for autonomous systems, and powerful new models—whether built for cyber tasks or general use—are already being evaluated and used within security agencies, creating both operational opportunities and safety tensions.
- The Pentagon requested about $54 billion for drone procurement and programs, a sum comparable to entire national budgets.
- The NSA is reported to be using Anthropic’s restricted Mythos model despite public friction between Anthropic and some defense stakeholders.
- Security researchers warn that powerful cyber‑oriented models can expose vulnerabilities faster than patches can be deployed.
What moved around the edges
Import AI flags automation in alignment research and HiFloat4 beating MXFP4 on Ascend chips
Jack Clark’s Import AI highlights work to automate parts of alignment research, reports a safety study of a Chinese model, and notes that Huawei’s HiFloat4 training format outperformed Western MXFP4 in an Ascend chip comparison.
Import AIGoogle rolls out three Ads Advisor features to speed up and harden ad workflows
Google has introduced three Ads Advisor enhancements intended to make ad creation and review faster and safer for advertisers, a rollout the company presents as a lead‑in to Google Marketing Live.
Google AI BlogEpic’s ‘Conversations’ tool lets creators add model‑driven NPCs to Fortnite
Epic is enabling creators to build AI‑driven characters on Fortnite islands with a new Conversations feature that replaces authored dialogue trees with model‑driven exchanges so players can interact directly with NPCs.
The Verge AIAmazon commits $5 billion and roughly 5 GW of custom silicon to Anthropic
Ars Technica reports Amazon invested $5 billion in Anthropic and allocated about 5 gigawatts of its custom silicon capacity to support rising demand for Claude models.
Ars Technica AIモデル開発で新しい論点が浮上
This week, during a podcast appearance, OpenAI CEO Sam Altman called out his competitor's new cybersecurity model, noting that the company was using fear to make its product sound more impressive than it actually is.
TechCrunch AI