In this episode

• OCC, FDIC, and Fed: Banks Can't Outsource Judgment to Algorithms — US banking regulators issued updated interagency model risk management guidance this week, moving from periodic attestation toward continuous, real-time oversight of AI and automated systems. Treasury released parallel AI risk management resources, and the OCC explicitly framed concentration in cloud and model providers as an interconnected systemic vulnerability requiring traceable, decision-level accountability.
• Mexico Advances Federal AI Law with Criminal Penalties and New Regulatory Institutions — Mexico's Senate committee, led by Senator Rolando Zapata, closed a 16-month drafting process involving 72 specialists and unveiled a federal AI bill that criminalizes non-consensual deepfakes, electoral manipulation, and autonomous lethal systems. It proposes a national AI strategy, a development fund, and a federal certification system — while drawing early scrutiny over vague terms like 'cognitive manipulation.'
• AI Omnibus Trilogue: High-Risk AI Deadline Likely Slipping to December 2027 — Parliament and Council agreed in principle to postpone the EU AI Act's Annex III stand-alone high-risk deadline from 2 August 2026 to 2 December 2027, with Annex I products sliding to August 2028. Political agreement is targeted for 28 April, formal adoption by July. The package also introduces a targeted prohibition on 'nudifier' apps and extends SME proportionality to small mid-caps. Until the Official Journal publishes the amendment, the August 2026 deadline remains binding law.
• US v. Heppner: Privilege and Work Product Don't Survive Casual AI Use — Judge Rakoff's bench ruling in US v. Heppner held that documents created with commercial generative AI and shared with counsel lose attorney-client privilege and work product protection if the AI tool doesn't maintain confidentiality — converting every prompt into a potential disclosure event.
• Madhya Pradesh HC: Section 11 Appointment Defects in ICA Cannot Be Cured by Consent — The Madhya Pradesh High Court voided ab initio an arbitral award in a dispute involving a Korean company on the ground that the arbitrator had been appointed by a High Court rather than the Chief Justice of India, as Section 11 of the 1996 Arbitration Act mandates for international commercial arbitrations. The court held that mandatory statutory provisions cannot be waived by party consent, conduct, or acquiescence.
• AI Is No Longer Borderless: Sovereignty Becomes a Contract Term — A deepset CEO analysis argues that borderless AI deployment is ending: the EU AI Act, India's DPDPA, Saudi cloud residency rules, and Microsoft's expanded Sovereign Cloud for disconnected AI are forcing enterprises to treat deployment location and model portability as core legal decisions. The piece uses the Anthropic–Pentagon dispute as a case study in vendor-lock geopolitical risk.
• DORA Is Live: Continuous Third-Party Oversight Becomes the Baseline — With DORA now fully operational, EU financial entities — and UK firms with EU operations — face continuous, evidence-based governance of ICT third parties rather than point-in-time audits. A January 2026 UK-EU MOU signals tighter convergence on vendor oversight, incident reporting, and sovereign data residency controls.
• German Social Courts Buckle Under AI-Generated Filings: 55% Spike in Emergency Proceedings — The Landessozialgericht NRW reports that AI-generated submissions to German social courts surged in 2025, with emergency proceedings up 55% to 7,615 cases. Judges describe the filings as verbose and legally incoherent, citing fabricated case law and straining benefits and unemployment dispute dockets.
• Iridius Raises $8.6M to Embed Compliance-as-Code into AI Workflows — Iridius, founded by alumni of Microsoft, AWS, and OpenAI, raised an $8.6M seed led by Chalfen Ventures with Accenture Ventures participating. The platform translates regulatory standards into executable logic embedded directly in enterprise AI workflows, initially targeting GxP-regulated life sciences with continuous compliance enforcement and automatic evidence generation.
• Forlex Prepares US Series A as LatAm Legaltech Routes Capital Offshore — Brazilian legal-AI startup Forlex is preparing a Q3 2026 US venture raise, with CEO Daniel Bichuetti relocating to California. The company is running a R$10–15M bridge from existing shareholders before the Series A, and has pivoted from gradual European expansion to direct US competition, leveraging an AWS partnership as institutional access.
• Mexico Reorganizes IP Leadership One Month Before USMCA Review — Santiago Nieto stepped down as IMPI director general to pursue a gubernatorial campaign, and all three judges of Mexico's Specialized IP Court (SEPI) departed simultaneously — one month before the July 1 USMCA review. New SEPI leadership, including Luis Edwin Molinar Rohana, may accelerate online litigation procedures.
• Ghana's Chief Justice: 'Technology Must Enhance Justice, Not Replace It' — At Ghana's National AI Strategy launch, Chief Justice Paul Baffoe-Bonnie warned against automating justice, insisting that AI remain subordinate to rule of law, constitutional values, and human judgment. He specifically flagged embedded bias, opacity, and accountability gaps in algorithmic decision systems.
• QBox: A Post-Quantum Theory Built on Causal Indefiniteness — Physicists James Hefford and Matt Wilson have developed QBox, a mathematical framework for a theory that could underlie quantum mechanics itself — much as quantum mechanics underlies classical physics. Its signature feature is causal indefiniteness: situations where the causal order between events is genuinely ambiguous, drawing a structural parallel to general relativity's treatment of spacetime.

Read the full briefing with sources →