Reporting from the edge of applied AI.Reporting from the edge of applied AI.

TLDR

Claim: valid JSON against a schema is enough to ship side-effecting agent tools. NIST Trustworthy and Responsible AI 800-5 (May 18, 2026) summarizes CAISI's agent-security RFI: commenters widely agreed agents present novel threats and fundamental cybersecurity practices require adaptation. The January 2026 RFI foregrounded indirect prompt injection and misaligned objectives. Verdict: overstated for workflows with side effects.

TLDR

NIST Trustworthy AI 800-5 summarizes RFI responses on AI agent security. Commenters widely agreed agents pose novel threats that hinder adoption and that core cybersecurity practice must adapt—not merely be reused. Government roles cited: implementation guidance, information-sharing, and standards. COSAiS SP 800-53 overlays remain forthcoming; prescriptive control IDs are UNKNOWN until NISTIR 8605D drafts.

TLDR

Colorado SB26-189 repeals and reenacts the SB24-205 AI consumer-protection provisions with automated decision-making technology (ADMT) rules for consequential decisions, the General Assembly bill page states. Developer documentation, deployer notices, a 30-day adverse-outcome description, three-year records, and a consumer path to human review are in the enacted summary. The attorney general must adopt rules on post-adverse-outcome disclosures by 1 January 2027. Skadden's 9 June 2026 client alert states Governor Polis signed the act on 14 May 2026. Independent confirmation of the signature date on a governor's press page is UNKNOWN.

TLDR

DeepSeek-V4-Pro and V4-Flash use interleaved Compressed Sparse Attention and Heavily Compressed Attention for 1M-token contexts. At 1M tokens the paper reports V4-Pro at 27% single-token FLOPs and 10% KV cache versus DeepSeek-V3.2; V4-Flash at ~10% FLOPs and ~7% KV. MIT weights on HuggingFace. Dollar cost per million agent tokens on your stack is UNKNOWN until benchmarked.

TLDR

McKinsey Global Institute estimates currently demonstrated technologies could in theory automate activities accounting for about 57 percent of US work hours — roughly 44 percent via agents and 13 percent via robots — while stressing this is technical potential, not a job-loss forecast. More than 70 percent of in-demand skills appear in both automatable and non-automatable work. The McKinsey Podcast episode published 30 April 2026 discusses human–agent hybrid teams. Named customer reorg outcomes are UNKNOWN.

TLDR

BIS published final rule 2026-06851 on 9 April 2026, effective 7 April 2026, extending the deadline to apply for Approved Integrated Circuit Designer status to 31 December 2026. The rule amends Note 1 to ECCN 3A090.a under the January 2025 advanced-computing due-diligence framework. Applicants received by that date may be treated as authorized IC designers for 180 days while BIS processes ERC review.

TLDR

OpenAI stated on 31 March 2026 that it closed a funding round with $122 billion in committed capital at an $852 billion post-money valuation. The 27 February 2026 post listed $110 billion in new investment at a $730 billion pre-money valuation, including $50 billion from Amazon, $30 billion from NVIDIA, and $30 billion from SoftBank. Enterprise share, token throughput, and revenue run-rate figures in those posts are company-reported. Independent audit of the round and of monthly revenue is UNKNOWN.

TLDR

Google Cloud documents a Gen AI evaluation service on Vertex AI: generate answers for a prompt set, score them with named rubrics, and log runs in Vertex AI Experiments. For RAG, operators use EvalTask datasets with prompt and response columns, then batch jobs for large golden sets. Groundedness and answer-quality metrics replace anecdotal pilots. Named customer before/after scores are UNKNOWN.

TLDR

Microsoft documents two Copilot agent paths — declarative agents using Copilot's orchestrator and models, and custom engine agents with external hosting — plus a Copilot Control System with security, management, and measurement pillars. Operators report choosing declarative for Graph-scoped Q&A and routing complex workflows to custom engines only after governance review. Named customer agent counts are UNKNOWN.

TLDR

Google's 19 February 2026 blog says Gemini 3.1 Pro is rolling out in preview for developers (Gemini API in AI Studio, Gemini CLI, Google Antigravity, Android Studio), enterprises (Vertex AI and Gemini Enterprise), and consumers (Gemini app and NotebookLM). Google reports a verified ARC-AGI-2 score of 77.1 percent, more than double Gemini 3 Pro. GitHub's changelog the same day puts 3.1 Pro in Copilot public preview for Pro, Pro+, Business, and Enterprise, with an admin policy gate for Business and Enterprise. Independent ARC-AGI-2 confirmation and a GA date are UNKNOWN.

TLDR

NIST's Center for AI Standards and Innovation announced the AI Agent Standards Initiative on 17 February 2026. CAISI and ITL will work on industry-led standards, open-source agent protocols, and agent security and identity research. Stakeholder input was open through CAISI's Request for Information on AI Agent Security, due 9 March 2026, and ITL's agent identity concept paper, due 2 April 2026.

TLDR

CAISI published a Request for Information on securing AI agent systems on 12 January 2026, docket NIST-2025-0035, with comments due 9 March 2026. The RFI focuses on threats distinct from conventional software: indirect prompt injection, poisoned models, and misaligned autonomous actions. NIST published summary analysis NIST AI 800-5 on 18 May 2026, reporting broad agreement that existing cybersecurity practice needs adaptation for agents.

TLDR

The Texas Legislature's enrolled summary for House Bill 149, the Texas Responsible Artificial Intelligence Governance Act, lists an effective date of 1 January 2026. The act amends the Business and Commerce Code and Government Code. It sets disclosure duties for certain governmental and health-care users of AI systems, prohibits specified harmful developments and deployments, creates civil penalties with Attorney General investigative authority, and establishes a regulatory sandbox and a Texas Artificial Intelligence Council. Section 8 of the enrolled analysis requires the Attorney General to post a specified website mechanism by 1 September 2026.

TLDR

California's Transparency in Frontier Artificial Intelligence Act (SB 53, Chapter 138) was approved 29 September 2025 and took effect 1 January 2026, per Business and Professions Code section 22757.12 as added by that chapter. A large frontier developer must write, implement, comply with, and clearly publish a frontier AI framework covering its frontier models. Before or concurrently with deploying a new or substantially modified frontier model, a frontier developer must publish a transparency report. The statute bars materially false or misleading statements about catastrophic risk. Whether a given lab's website posting satisfies the statute is UNKNOWN without comparing that posting to the code.