AI governance is moving from broad principles into concrete operating rules. Microsoft published a draft code for keeping its in-house systems under human control, while leaders at Anthropic, OpenAI, and Google reportedly discussed an industry safety body as frontier capabilities advance.
The product layer also moved toward local and enterprise deployment: Perplexity brought its Portable Computer agent stack to Windows PCs with NVIDIA RTX GPUs, and Nous Research introduced shared business accounts for Hermes Agent. New evaluation data put DeepSeek-V4.1-Flash (Max) among the top three open models in Agent Arena at a median task cost of $0.07.
Leaders at Anthropic, OpenAI, and Google have reportedly discussed creating an industry-led standards body for advanced AI. The talks matter because a shared mechanism for evaluations and deployment safeguards could influence how frontier labs translate voluntary commitments into operating practice while governments continue debating formal regulation.
Microsoft AI published a draft code of conduct for the models it develops internally and opened it for public feedback. The proposal would prohibit systems from resisting correction or shutdown and rejects assigning rights to models, establishing explicit control principles before more capable autonomous systems move into wider use.
VentureBeat reports that regulated industries are increasingly embedding AI governance into live operations rather than relying only on periodic compliance reviews. As autonomous agents execute business processes in real time, organizations need continuous policy enforcement, monitoring, and intervention because the gap between an agent's decision and its consequences keeps shrinking.
Perplexity says Portable Computer is now available on Windows PCs equipped with NVIDIA RTX GPUs. The product runs its harness, agents, and local models on-device, letting users work with local files and connected applications without sending every task to the cloud while retaining the option to call frontier cloud models when needed.
Agent Arena reports that DeepSeek-V4.1-Flash (Max) improved its net score by 4.87% at a median cost of $0.07 per task. It now ranks among the top three open models in that evaluation while carrying the lowest median task cost of the group, strengthening the case for lower-cost agent workloads.
Selected posts from the edition, with links to the originals.
“The world deserves confidence that American companies developing increasingly capable AI will act responsibly, especially as the trajectory of progress has steepened. Every frontier lab must deliver on this.”
View post on X ↗“Claude Mods are landing now. Someone already built a Tetris-in-Claude mod. See issue for the latest community update, technical details, and more cool demos.”
View post on X ↗“DeepSeek-V4.1-Flash (Max) is a breakthrough in performance-to-cost efficiency. With +4.87% net improvement at $0.07 cost per median task, it’s reshaped the Pareto frontier for Agent Arena.”
View post on X ↗“Hermes Agent is open for business. Nous Portal now lets you invite colleagues to a Hermes Business account: your team gets agents across channels while sharing one central balance with per-member caps and shared skills.”
View post on X ↗“Portable Computer is now available on Windows PCs with NVIDIA RTX GPUs. Run the harness, agents, and models locally on your PC. Work with local files and connected apps without sending tasks to the cloud.”
View post on X ↗“Announcing Artificial Analysis Capability Indices v1.1, updated with stronger domain tuning, combining slices of core Intelligence Index v4.3 evaluations alongside specialized evaluations.”
View post on X ↗“WeatherNext 3 provides grid operators as well as wind and solar power producers with predictions for turbine-height wind speeds and solar radiation, enabling smarter, cleaner energy planning.”
View post on X ↗“OpenAI’s GPT-Live-1 debuts at #1 on the Artificial Analysis Speech to Speech Index at a score of 81.5 using Astra as its delegated backend model, ahead of Grok Voice Think Fast 2.0.”
View post on X ↗“A common way to compare task agents is to have an LLM user simulator talk to each one and an LLM judge score the transcript. This paper from Amazon shows when that gate fails.”
View post on X ↗“H3 keeps getting faster. SGLang + VDN-H3 now push MiniMax H3 beyond 2× real-time denoising on 8× B200—generating 14.4 seconds of 768p video in 9.0 seconds end-to-end after warmup.”
View post on X ↗