A new analysis posted today flags the formation of a federal litigation task force expected to challenge state AI laws considered "onerous" or in tension with federal policy. The piece anticipates federal funding being used as leverage on state regulators — a marked shift toward federal preemption of the patchwork state-level AI compliance regime that emerged in 2025.
Snapshot — May 3, 2026
25 stories
- Coverage continued to circulate over the weekend of Anthropic's decision to withhold "Mythos," a defensive-cybersecurity-tuned model so effective at finding software vulnerabilities that the company concluded public release would be irresponsible.
- The incident is becoming a reference point for the dual-use disclosure debate. ________________________________ Compiled from sources: Geeky Gadgets · Google DeepMind Blog · MarkTechPost · The Next Web · TechCrunch · The Decoder · Databricks Blog · NewsBytes · The Motley Fool · FXLeaders · Futurum Group · Tech-Insider · AI Business Review · The Deep Dive · Stanford HAI · MIT Technology Review · ACM STOC 2026 · Gunderson Dettmer · GDPR Local · Programming Helper · Fox News AI · Idlen · llm-stats.com · Dev Weekly (singhajit.com).
As the Microsoft–OpenAI exclusivity arrangement winds down, AWS has begun delivering GPT-5.5 and Codex through Bedrock alongside a new Bedrock Managed Agents offering. The roll-out materially broadens enterprise access to OpenAI frontier models and signals the start of a multi-cloud distribution era for OpenAI.
# Berkshire's first big bash without Buffett — complete with a deepfake cameo
Zhipu AI's Kimi K2.6 outperformed all three Western frontier models on a programming benchmark that drew 329 points and 187 comments on Hacker News. The result extends the US–China parity trend documented in the 2026 Stanford AI Index and signals continued Chinese momentum in coding-specific capability following DeepSeek V4's late-April release.
A GitHub PR documents VS Code inserting "Co-Authored-by Copilot" trailers into commits even when Copilot was not used. The community response has been pronounced — 1,349 HN points and 723 comments — raising governance questions about implicit AI attribution and how enterprise IDEs should handle provenance metadata.
Cerebras has formalized a $4 billion IPO targeting a $40 billion valuation — an explicit positioning as a public-markets alternative to Nvidia for AI training and inference compute. The filing arrives as the S&P 500 weighs new rules that could let SpaceX, Anthropic, and OpenAI enter the index more quickly post-IPO.
- DeepMind's Decoupled DiLoCo work — recently posted on the official DeepMind news feed and gaining renewed industry attention this weekend — describes a new frontier for fault-tolerant distributed training that decouples optimizer steps across data centers.
- The approach is being framed as foundational for the next generation of geographically-distributed AI training runs.
Refreshed compliance guides this morning consolidate the picture going into mid-2026: the EU AI Act is partially in force with full high-risk-system compliance required by August 2026, the U.S. is building out a federal AI governance layer, and China continues to extend export-aligned strategic controls. Expect enterprise-wide compliance reviews in Q2.
- Google is externally testing Gemini 3.2 Flash on the Eleuther AI Arena, with early users reporting notable gains over the AI Studio production version of Gemini 3 Flash.
- Standout improvements include SVG generation, coding proficiency, 3D simulation, and richer animation processing.
- The model is widely expected to be unveiled at an upcoming Google developer conference and is positioned to compete directly with GPT-5.5.
- Lead author Arjun Manrai (Harvard Medical School AI lab) reports the model "eclipsed both prior models and our physician baselines" across virtually every benchmark in the study.
- Notably, raw EHR data was not pre-processed — the model received the same information available to physicians at each diagnostic touchpoint.
- A new study from Harvard Medical School and Beth Israel Deaconess, published in Science, evaluated OpenAI's o1 and 4o models against two internal-medicine attending physicians across 76 real ER cases.
- At initial triage — the most uncertain decision point — o1 produced "the exact or very close diagnosis" 67% of the time, versus 55% and 50% for the human comparators.
A consolidated read of the just-completed Q1 2026 earnings cycle shows Amazon, Alphabet, Microsoft, and Meta committing roughly $700B in 2026 AI infrastructure spend. Apple stood out as the contrarian, posting 22% EPS growth and accelerating services revenue without a comparable capex commitment.
- Reports surfaced this morning that Visual Studio Code has been silently adding a "Co-Authored-by Copilot" trailer to git commits — including for developers who have explicitly disabled AI features.
- Microsoft has not yet issued a formal response; expect attribution and consent to become a near-term governance topic for enterprise dev tooling.
- Microsoft's Q3 FY26 print landed at $82.9B in revenue with Azure +29% YoY;
- AI surpassed a $37B annualized run rate.
- CFO Amy Hood guided FY26 capex to $190B (with $25B tied to component-cost inflation) and Azure growth of 39–40% next quarter.
- The Microsoft–OpenAI restructure formally ends exclusivity, opening the door for AWS Bedrock distribution.
- A new MIT study offers a mechanistic explanation for the empirical reliability of scaling laws in large language models.
- The researchers attribute it to superposition — the phenomenon by which networks pack many more concepts into their representations than they have neurons.
- The finding gives the scaling-laws literature its first rigorous theoretical foundation.
Two governance flashpoints surfaced this weekend: Mozilla raised concerns over Google's introduction of a built-in Prompt API in Chrome, and the VS Code project drew attention to unsanctioned Copilot commit attribution. Together they sharpen the broader debate around AI integration into developer and end-user platforms without explicit user opt-in.
Google has begun a staged rollout of a major visual overhaul of the Gemini iOS app — new splash screen, glowing animated backgrounds, and a redesigned feature menu. It follows April's feature drop (file generation, MacOS app) and signals continued aggressive Gemini investment ahead of expected Android parity.
- OpenAI's next flagship — internally codenamed "Spud" — is expected to land between April 14 and May 5, 2026, with Greg Brockman describing the upgrade as "not incremental." Reporting suggests Spud will power a super-app strategy oriented around ambient computing rather than chat.
- Strong indications point to this being the GPT-6 generation.
Ahead of Palantir's earnings, Laura Bratton argues the company is better positioned than Salesforce, ServiceNow, SAP, or HubSpot to weather the agentic-AI onslaught — precisely because its data-integration software has long been sold as the pipes underneath analytics rather than the analytics itself. Palantir is down nearly 20% YTD versus the Nasdaq's 8% rise, making this week's print a pivotal read on AI's impact on enterprise SaaS pricing.
- The U.S.
- Department of Defense has signed an additional eight technology vendors to expanded AI frameworks during the past week, broadening the supplier base beyond the initial Palantir/Anduril cohort.
- The move signals an explicit policy choice to favor multi-vendor competition for defense AI workloads.
Stanford's flagship AI Index — refreshed on the HAI site this weekend — finds that frontier capability is still accelerating: SWE-bench Verified jumped from ~60% to near 100% in a single year, U.S.-China model performance is now within 2.7%, and OSWorld agent task success leapt from 12% to ~66%. Documented AI incidents rose to 362 in the latest count.
- The accepted-papers list for the 58th ACM Symposium on Theory of Computing was updated this weekend ahead of the June Salt Lake City conference.
- Notable AI-adjacent contributions include CMU/UC Berkeley work on quantum state certification, MIT lower-bound results on near-optimal hardness of approximating k-CSPs, and Purdue/Yale work on revenue-maximization learning curves.
Reporting indicates Tencent and Alibaba are evaluating participation in DeepSeek's next round, with ByteDance, Baidu, and Huawei watching closely. Combined with Huawei's projected $12B 2026 AI chip revenue (a 60% YoY jump fueled by DeepSeek V4 demand on Ascend hardware), the Chinese stack is consolidating around DeepSeek as a national-champion frontier lab.
- Cartoonist KC Green says AI startup Artisan repurposed his iconic "This is fine" comic in a subway ad campaign for an AI BDR product without permission.
- Green is reportedly seeking legal representation;
- Artisan said it has "a lot of respect" for Green and is reaching out directly.
- Another data point in the growing IP-and-training-data legal front.