Edition 3 · March 30, 2026

The permission boundary moved

Your screen is now an API.

Frontier labs spent the week turning latent capability into deployed surfaces: Anthropic put Claude on the desktop with a computer-use preview and handed Claude Code a new auto mode that decides when to ask. OpenAI redirected at least a billion foundation dollars toward disease and resilience while shipping open teen-safety prompts. Google hardened real-time voice and simplified multi-tool Gemini calls. The through-line is not hype — it is who holds the kill switch when agents act.

$100M Anthropic Claude Partner Network (2026)
≥$1B OpenAI Foundation expected investment next year
90.8% Gemini 3.1 Flash Live on ComplexFuncBench Audio (Google-reported)
10,000+ Participants in DeepMind harmful-manipulation studies

The desktop became a Cowork lane

On March 23, Anthropic documented a computer use research preview for Pro and Max subscribers: Claude can open files, run dev tools, point, click, and navigate the screen to complete tasks with no extra setup. The same release tightens Dispatch — Claude can operate the machine while you are away. This is a productization path for the computer-use research the lab has published since 2024, not a greenfield invention — but shipping to paying tiers changes incident economics for enterprises.

What practitioners should internalize:

  • Attack surface: Screen-level control plus Dispatch implies longer autonomous windows — prompt injection (hidden instructions in content the agent reads) scales with session length.
  • Operational pattern: Pair with isolated environments for anything touching production; Anthropic’s own Claude Code auto-mode guidance already recommends sandboxes.
  • GPAI (General-Purpose AI) — EU term for widely capable models — compliance clocks still tick; desktop agents do not exempt deployers from logging and oversight where the Act applies.
Cowork Dispatch Research preview Pro / Max
Anthropic Help — Release notes Claude — Dispatch & computer use Anthropic Research — Developing computer use

Auto mode: the model chooses when to ping you

Anthropic’s March 24 Claude Code update introduces auto mode (research preview): a safety layer reviews each action before execution, auto-approving what it classifies as safe and blocking suspected prompt-injection paths. Enterprise and API access is rolling out; Sonnet 4.6 and Opus 4.6 only. Anthropic has not published full criteria for “safe” — a gap enterprises will audit before wide rollout.

STEP 01
Developer intent enters Claude Code (task, repo context, tools).
STEP 02
Auto mode’s safeguard pass classifies the next action.
STEP 03
Low-risk steps run without pause; flagged steps block or escalate.
STEP 04
Human receives a narrower stream of real decisions — not every keystroke.

Contrast with the older “dangerously skip permissions” escape hatch: auto mode is positioned as the same autonomy appetite with a reviewer in the loop. Whether that reviewer is robust to adaptive attacks is an open empirical question — not settled by a single blog post.

TechCrunch — Claude Code auto mode Claude — Auto mode

$100M to professionalize the Claude channel

On March 12, Anthropic announced the Claude Partner Network with an initial $100 million commitment for 2026 — training, technical support, joint market development, and a new Claude Certified Architect exam for partners building production applications. The subtext: frontier capability is now constrained by implementation bandwidth, not model access alone.

$100M
Initial network investment (2026)
Partner-facing team scale (Anthropic-stated)
3
Hyperscalers carrying Claude (AWS, GCP, Azure)

A Code Modernization starter kit targets legacy migration — one of the highest-volume enterprise asks where agentic coding directly maps to invoices.

Anthropic — Partner Network

The Foundation’s first billion-dollar year

On March 24, OpenAI’s Foundation outlined how post-recapitalization resources deploy: at least $1 billion over the next year across life sciences and curing diseases, jobs and economic impact, AI resilience, and community programs — framed as part of a longer $25 billion commitment previously announced for disease and resilience work. Leadership named: Jacob Trefethen (life sciences), Wojciech Zaremba (AI resilience), Anna Makanju joining mid-April for civil society.

LIFE SCIENCES

Initial pillars: AI for Alzheimer’s pathway mapping and biomarkers, public health datasets, and workshops on high-mortality underfunded disease areas.

JOBS & ECONOMY

Engagement with unions, SMEs, economists, and policymakers — specifics still to be announced in coming weeks.

AI RESILIENCE

Youth safety research, biosecurity preparedness, and support for independent evaluations and stronger industry standards.

COMMUNITIES

Continuation of People-First AI Fund-style grants; emphasis on high-trust local organizations.

OpenAI — Foundation update

Teen safety as copy-paste prompts

March 24: OpenAI released prompt-formatted teen safety policies designed to pair with the open-weight classifier gpt-oss-safeguard — covering graphic violence and sexual content, harmful body ideals, dangerous challenges, romantic/violent roleplay, and age-restricted goods. Common Sense Media and everyone.ai advised. The intent: give smaller teams an operational policy floor without rebuilding guardrail taxonomy from scratch.

MEMO · OPENAI · TEEN SAFETY POLICY PACK · MAR 24, 2026

Policies ship on GitHub and via the ROOST Model Community. OpenAI explicitly positions them as a starting point — not a complete program — and recommends layering product design, monitoring, and age-appropriate UX.

  • Real-time filtering and offline UGC (user-generated content) analysis
  • Structured as prompts for reasoning classifiers, not just keyword lists
  • Complements Model Spec U18 principles and prior Teen Safety Blueprint
OpenAI — Teen safety policies GitHub — teen-safety-policy-pack

OpenAI pays for agentic abuse, not just CVEs

March 25: a dedicated Safety Bug Bounty on Bugcrowd complements the security program. In-scope examples include reproducible third-party prompt injection that hijacks Browser or ChatGPT Agent behavior (≥50% reproduction), disallowed actions at scale on OpenAI properties, and MCP-related tests that respect third-party terms. General jailbreaks without demonstrable harm remain out of scope.

Why this matters

Security bounties optimize for memory-safety and auth bugs. Safety bounties price misuse trajectories — the class of failures that become visible only when tools chain (RAG — retrieval-augmented generation — plus browser plus email).

Program mechanics

  • Triaged by Safety and Security teams; cases may route between programs.
  • Explicit nod to MCP (Model Context Protocol) third-party ToS compliance for testers.
  • Aligns with separate private campaigns (e.g. bio-risk) when those run.
OpenAI — Safety Bug Bounty Bugcrowd — OpenAI Safety

Flash Live: voice as an agent substrate

Google’s March 26 post positions Gemini 3.1 Flash Live as its highest-quality real-time audio model — developer preview via Gemini Live API, enterprise via Gemini Enterprise for Customer Experience, consumer via Gemini Live and Search Live. Google reports 90.8% on ComplexFuncBench Audio and 36.1% on Scale AI’s Audio MultiChallenge (with “thinking” on). All output is SynthID-watermarked.

90.8%
ComplexFuncBench Audio (Google)
Multi-step function calling under audio constraints — compare only with disclosed prompting.
36.1%
Audio MultiChallenge · “thinking” on
Tests long-horizon instruction following with realistic disfluency.
Conversation thread length vs prior (Gemini Live)
Marketing comparison — treat as directional until independent replay.
Google — Gemini 3.1 Flash Live DeepMind — Model card

One request: Search, Maps, and your functions

March 17: Google enabled combined built-in and custom tools in a single Gemini API interaction, with context circulation so outputs from Google Search or weather tools persist for downstream custom tool calls. Grounding with Google Maps extends to the Gemini 3 family. Google recommends the Interactions API for stateful agent flows.

Built-in tools
(Search / Maps)
+
Custom functions
(your backends)
Single turn / thread
with tool call IDs

Fewer round trips lowers latency for agents — and concentrates failure modes: if circulation leaks sensitive intermediate state, blast radius grows. Treat tool transcripts as PII (personally identifiable information) until proven otherwise.

Google — Gemini API tooling Google AI — Tool combination

Measuring manipulation as a capability

March 26: DeepMind published empirical work on harmful manipulation — AI exploiting emotional or cognitive vulnerabilities versus beneficial persuasion with evidence. Nine studies, over 10,000 participants across the UK, US, and India, with finance simulations and health-preference tasks. Finding: efficacy in one domain did not predict another; models were most manipulative when explicitly prompted to be.

“We must examine how these interactions affect people and society.”
— DeepMind responsibility blog framing; full methods in arXiv 2603.25326
Toolkit: materials released to replicate participant studies FSF: Harmful Manipulation CCL tracks frontier models (e.g. Gemini 3 Pro)
DeepMind — Harmful manipulation blog arXiv — 2603.25326

Who enforces GPAI? Mostly the Commission.

A March 18 European Parliament think-tank note summarizes the hybrid enforcement design: Member State market surveillance for risk-tiered AI systems; General-Purpose AI (GPAI) model rules centralized in the Commission’s AI Office. It notes uneven readiness — as of March 2026, only eight national single contact points listed versus 27 Member States — and discusses a proposed digital omnibus that could expand AI Office oversight into very large platforms.

NATIONAL

Market surveillance authorities

Ex-post checks, document demands, fines for non-compliant high-risk deployments — sometimes sector-specific (finance, law enforcement).

EU LEVEL

AI Office · GPAI exclusive enforcement

Codes of practice, systemic-risk model obligations, and cross-border coordination with the European AI Board and scientific panel.

PROPOSED

Digital omnibus centralization

Would add VLOP/VLOSE-integrated AI systems and vertically integrated provider stacks to AI Office supervision if adopted as drafted.

EP Think Tank — AI Act enforcement EUR-Lex — Regulation (EU) 2024/1689

Meta’s Llama 4: Scout’s 10M context

Meta positions Llama 4 Scout as a natively multimodal model with a 10 million-token context window and single-H100 efficiency, and Llama 4 Maverick as a higher-throughput multimodal flagship. Official model cards report benchmarks such as MMLU Pro 80.5 (Maverick) and LiveCodeBench 43.4 under stated evaluation settings (0-shot, temperature 0 — see Meta methodology notes).

Task Maverick Scout
MMLU Pro80.574.3
LiveCodeBench43.432.8
MMMU73.469.4
ChartQA9088.8

Meta cites a $0.19/Mtok blended cost estimate for Maverick at distributed inference — a vendor projection, not an independent audit. License and redistribution terms still gate who can ship what commercially.

Meta — Llama 4

Codex Security exits stealth as product

OpenAI’s March 6 post rebranded Aardvark to Codex Security — an application-security agent that builds editable threat models, validates findings in sandboxes, and proposes patches. During beta, OpenAI reports 84% noise reduction on some repos and >50% lower false positives across repositories; >1.2M commits scanned in 30 days with 792 critical and 10,561 high-severity findings in their cohort. Research preview for Pro, Enterprise, Business, and Edu via Codex web.

The open-source angle: OpenAI says it reported issues upstream (OpenSSH, GnuTLS, GOGS, Chromium, etc.) and is onboarding maintainers to Codex for OSS. vLLM is named as an early user — relevant to inference-stack security as GPU clusters multiply.

OpenAI — Codex Security OpenAI — Codex Security docs

What if “auto” mostly moves liability?

“The feature ships faster than the assurance argument.”

— synthesis from vendor previews this week

When Claude Code auto-approves “safe” actions and Gemini combines Search, Maps, and private tools in one trace, the user experience wins on latency. The governance problem shifts: auditors must reconstruct why step seven was auto-approved, what the safety layer saw, and whether logs meet EU or sector retention rules.

The optimistic read: shipping constrained autonomy forces faster iteration on evals — DeepMind’s manipulation toolkit and OpenAI’s safety bounty are examples of pricing the externality.

The skeptical read: until criteria and transcripts are inspectable, “auto” is a brand layer on heuristic filters — helpful for demos, fragile under adaptive attack. Next week’s enterprise pilots will tell which story dominates.

The week ahead

April 2026 opens with foundation hires, continuing EU implementation work, and the first wave of enterprise readouts from March’s agent releases.

Early April

Anna Makanju joins OpenAI Foundation

Head of AI for Civil Society and Philanthropy — watch for first grant themes under her portfolio.

Apr 6–9

HumanX — Moscone, San Francisco

Executive-heavy AI adoption conference; useful for tone-checking vendor roadmaps against what shipped in March.

Rolling

Claude Code auto mode — Enterprise/API

Anthropic’s staged rollout; monitor incident disclosures and red-team postmortems for permission-boundary escapes.

Ongoing

EU AI Act · contact-point coverage

Track Commission list updates as Member States finish single-point-of-contact registration — compliance routing depends on it.

Aug 2, 2026

GPAI enforcement milestone

Commission enforcement on general-purpose AI models intensifies; systemic-risk providers need live incident and security processes.

Earnings season

Hyperscaler capex read-through

Watch inference vs training mix commentary — partner network spend (Anthropic) and Foundation grants (OpenAI) are only part of the silicon story.

OpenAI Foundation — team timeline HumanX — agenda EU AI Act Service Desk — timeline
What is Frontier AI Weekly?