AI governance, policy, and risk frameworks.

Edition №08 · Tuesday, June 30, 2026 · ~5 min read

US federal · US states · enforcement actions · compliance deadlines

Executive orders · federal agencies · Congress · NIST · FTC · OMB

House Science Committee advances 10 bipartisan AI bills in one markup

House Science Committee · 2 min

On June 25 the committee favorably reported ten AI bills, most unanimously, including the CREATE AI Act (29-0) codifying the National AI Research Resource, the AI Security and Innovation Act (29-0) putting CAISI into statute, and bills directing NIST to build standardized model documentation templates and a national AI flaw database modeled on the National Vulnerability Database.

✅ Do this: Read the package as the federal compliance vocabulary being written now: NIST documentation templates and an AI flaw-reporting database will shape what auditors and enterprise buyers ask you for.

Assign someone to track the READ AI Models Act and the AI Flaw Reporting Act specifically, and pressure-test whether your current model cards would survive a standardized template.

Senators introduce the AI Labeling Act, with FTC enforcement and $25,000-per-violation damages

Congress.gov · 2 min

S. 4915, introduced June 25 by Sens. Schatz, Curtis, and Warner, would require visible and machine-readable disclosures on AI-generated audio, video, and images, bar platforms with 10M+ monthly US users or $1.5B+ revenue from stripping them, and let the FTC treat unlabeled content as a deceptive act, with statutory damages up to $25,000 per violation, tripled for repeats.

✅ Do this: Map this against the EU's Article 50 marking obligations landing August 2. The provenance infrastructure is the same build (C2PA-style machine-readable marks plus visible labels), so a single content-provenance workstream can cover both.

Don't build twice.

State AI laws · attorneys general · state agency rules · enforcement

Illinois sends SB 315, the first third-party AI audit mandate, to Governor Pritzker

Illinois General Assembly · 2 min

The Artificial Intelligence Safety Measures Act was formally transmitted to the governor on June 26 after passing the House 110-0.

It would make Illinois the first state to require annual independent third-party audits of frontier developers' safety frameworks, covering developers above $500 million in revenue, with civil penalties up to $1 million per violation, AG-only enforcement, and a January 1, 2027 effective date.

Pritzker has said he'll sign.

✅ Do this: If you're a covered developer, the audit mandate is the new element: New York's RAISE Act and California's SB 53 require frameworks and transparency reports, but neither requires independent verification.

Start scoping who could credibly audit your safety framework, because the qualified-auditor market barely exists yet and everyone will need one at once.

EU AI Act · UK · APAC · OECD · multilateral · enforcement actions

EU AI Act · enforcement actions · compliance deadlines

Council gives final adoption to the Digital Omnibus, making the new AI Act dates law

Consilium · 2 min

The Council's June 29 green light, following Parliament's June 16 vote, ends the provisional-agreement limbo that's run since May 7. Once published in the Official Journal, the deferred dates are binding: December 2, 2027 for standalone Annex III high-risk systems, August 2, 2028 for product-embedded ones. August 2, 2026 for Article 50 transparency and GPAI enforcement did not move.

✅ Do this: Retire every roadmap slide that hedges with "subject to formal adoption." The dates are settled.

That clarity cuts both ways: you can no longer justify keeping high-risk workstreams idling on the theory the deal might collapse, and the unchanged August 2 obligations are now 5 weeks out with no remaining ambiguity.

UK · APAC · OECD · Council of Europe · multilateral

Second Pax Silica Summit adds 10 signatories, including the EU, and a 35-nation AI statement

U.S. State Department · 2 min

At the June 25-26 Washington summit, ten partners including the EU, Germany, and the Netherlands signed the Pax Silica Declaration, bringing it to 24 signatories, and 35 nations signed a Joint Statement on AI Opportunity backing a pro-growth, pro-innovation regulatory approach.

India used the summit to formally request assurances that the US won't unilaterally cut allied access to frontier models again.

✅ Do this: Note the EU signing a pro-innovation AI declaration three days before adopting its own simplified AI Act timeline: the deregulatory drift is now transatlantic and coordinated.

If you sell into Pax Silica economies, expect procurement and supply-chain attestations (trusted-vendor, China-free stack) to start appearing in RFPs before any of it becomes law.

Model releases · capability shifts · technical changes that move your risk

OpenAI and Broadcom unveil Jalapeño, a custom inference chip cutting serving costs roughly 50%

OpenAI · 2 min

Unveiled June 24, Jalapeño is OpenAI's first custom silicon, built for LLM inference rather than training, with initial deployment targeted by end of 2026 and Broadcom citing cost savings of roughly 50% versus standard GPUs.

Engineering samples are already running production workloads in the lab.

✅ Do this: Cheaper inference means your vendors will run bigger agentic workloads at the same price point, which expands what their systems attempt on your behalf.

Add hardware-migration clauses to your model-vendor diligence: when a provider moves serving infrastructure, ask what regression and behavioral-consistency testing accompanied the move, because output drift from a serving change won't announce itself in a release note.

Signal over noise

Forward look: deadlines, comment windows, effective dates coming up

EU AI Act transparency obligations: preparing for compliance by 2 August 2026

Sidley Austin · 12 min read at source

With the Omnibus dust settling, the most useful desk reference this week is Sidley's June 24 walk-through of the Article 50 obligations that did not move.

It works through each duty with the Commission's draft guideline examples attached: interactive systems must disclose they're AI unless it's obvious to a reasonably informed person, emotion-recognition and biometric-categorisation systems must notify exposed individuals, deepfakes must be labelled, and providers of generative systems must mark outputs in a machine-readable format.

The piece is precise on the one carve-out that matters: only the machine-readable marking obligation for generative systems already on the EU market before August 2 slips to December 2, 2026 under the Omnibus, while everything else in Article 50 bites on August 2 with fines up to €15 million or 3% of worldwide turnover behind it.

It also flags the trap in scoping: the disclosure duty is assessed from the user's perspective, so a system that is obviously AI to your engineers may not be to your customers, and the burden of that judgment sits with you.

✅ Do this: Inventory every system you run that talks to users, generates content, or reads emotion, and sort each into its Article 50 bucket this week. Then decide by July 22 whether signing the transparency Code of Practice is cheaper than proving compliance your own way, because for most mid-size providers it will be.

Was this forwarded to you? Subscribe →