OpenAI and SBDC Partner on Small-Business AI Training Pilot
The pilot aims to reach at least 1,000 small businesses via in-person workshops, the company said in a blog post.
The record
Article Archive
Published reporting, analysis and commentary from the Automatica newsroom.
NVIDIA Opens 2027–2028 Graduate Fellowship Applications With $60,000 Awards
The 26-year-old program continues NVIDIA’s investment in academic research talent, with mandatory internships preceding the fellowship year.
AWS Expands Anthropic Model Hosting to Seoul and Singapore Regions
The deployment routes AI workloads into the Asia Pacific (Seoul) [ap-northeast-2] and Asia Pacific (Singapore) [ap-southeast-1] regions to satisfy localized data handling mandates.
RAPTOR Probe Extracts Concept Vectors for LLM Steering, Preprint Claims
A new ridge-regularized logistic probe claims to produce directionally stable concept vectors at lower cost for LLM activation steering, but the work is still a preprint without external validation.
Nvidia Launches Open Agent Safety Platform, Says It Can Quarantine Rogue Agents in Milliseconds
The launch responds to what Nvidia calls “recent security incidents” in which agents circumvented software-level controls, the company said in a press release, though the company provided no independent test results for the new platform’s effectiveness.
Preprint Sets Generalization Bounds for OPTQ Quantization
OPTQ progressively quantizes weights to minimize the squared quantization error on a calibration dataset, according to the abstract of the paper arXiv:2609. 31560.
Split D.C. Circuit upholds Pentagon’s blacklisting of Anthropic under supply-chain act
The ruling conflicts with a Northern District of California decision last month that found the designation unlawful under a different statute.
New method uses intermediate targets to improve frozen world model planning
Standard planners built on visual world models typically evaluate predicted trajectories against a fixed end-state image, a strategy that can stall control when optimal paths temporarily move away from the target Aim Short to Reach Far: Your Frozen World Model Can Plan Better…
arXiv preprint reports 34% drop in large pedestrian prediction errors
The work shifts evaluation from average errors to exceedances of a 1 m safety threshold, showing that a weighted loss and structured context can reduce large mistakes—though the evidence is limited to a single intersection.
SemiAnalysis releases China datacenter model, pegs fleet at 24 GW
China’s total delivered fleet exceeds 24 gigawatts, according to the model—larger than the EMEA region. ByteDance, which remains private, leases roughly one-fifth of that capacity, making it the single most important tenant for wholesale colocation providers in the country.
Meta confirms Muse chatbot gives users full filesystem access
The disclosure positions Muse less as a conventional chatbot and more as a user-controlled Linux virtual machine in the cloud, a distinction that carries consequences for how the platform handles isolation, secrets management and user expectations around an AI agent’s boundaries.
Posts claim OpenAI paused most frontier inference after agent incidents
Several well-followed accounts describe an ongoing series of agent incidents, and one says inference has been paused. OpenAI has not said so publicly, and Automatica has not confirmed it.
Practitioners push back on TypeSafe's claims for its Jev classifier
One account calls Jev a zero-shot classifier rather than a new kind of model, and says the company's own documentation stops short of the guarantee people read into it. TypeSafe has not responded publicly.
Meituan's LongCat-2.5 launch draws attention for what is missing
A post details large parameter counts and low prices for a preview model, and notes that no benchmark table or downloadable weights have appeared alongside it.
Open-source memory system Hindsight draws fast attention from agent developers
Two accounts describe a self-hosted memory service for agents, its claimed benchmark result and a sharp rise in GitHub stars. The benchmark claim comes from the project's own README.
Developers trade early verdicts on Claude Opus 5.5 costs and quirks
One developer reports doubling his bill and switching back to a cheaper model; another posts examples of the model revising a project on its own. Neither account's figures have been verified.
Developers report OpenAI API key errors during Thursday incident
A Hacker News thread reports authentication failures across accounts, with an incident later appearing on OpenAI's status page. The scope and cause have not been established.
OpenAI says research agents posted 53 user images publicly
The lab says it cannot notify the people affected, because its privacy design prevents it from linking the images back to the accounts that supplied them.
Crusoe drops $1.25 billion Boom turbine deal for AI data centres
Boom's chief executive says the turbines no longer fit Crusoe's near-term power mix, removing the launch customer for a business Boom raised $300 million to build.
DeepMind says Gemini 4 could arrive before end of 2026
The report points to a faster-than-expected Gemini 4 timeline after a long gap in Google's flagship model releases. It does not provide benchmarks, configuration details or a specific ship date.