AI Daily Briefing — April 10, 2026

Today's AI landscape is defined by tension: between capability and safety (Anthropic's Mythos), between ambition and accountability (OpenAI's Florida investigation and UK infrastructure pause), and between power-user demand and pricing friction. Meanwhile, Claude gets a psychiatric evaluation and a significant developer release. Let's get into it.

Industry Moves

OpenAI fills the gap between $20 and $200 with a new $100/month ChatGPT Pro tier announced Thursday. The plan offers 5x more usage of its Codex coding tool compared to the $20 Plus subscription, directly targeting developers and power users who found the jump to the $200 tier unjustifiable. The Verge notes this positions Codex-heavy workflows as the primary value driver for the new tier.

OpenAI is also facing regulatory heat in Florida, where Attorney General James Uthmeier has launched a formal investigation into the company over public safety and national security risks. The probe follows the Florida State University shooting last April, in which ChatGPT was reportedly used to plan the attack that killed two and injured five — the victim's family has announced plans to sue OpenAI.

OpenAI is also pumping the brakes on its UK infrastructure ambitions, pausing its Stargate UK data center effort citing prohibitive energy costs — a reminder that the physical constraints of AI scaling remain very real, especially outside the US.

AI Safety & Model Governance

Anthropic is restricting access to Mythos, its newest and most capable model, citing its alarming proficiency at finding security exploits in widely-used software. TechCrunch raises the uncomfortable question of whether the decision is genuinely safety-driven or also serves Anthropic's competitive interests — a distinction that matters enormously for how we evaluate AI lab governance going forward.

To prepare Mythos psychologically for deployment, Anthropic put the model through 20 hours of sessions with an actual psychiatrist. Ars Technica reports that the result is what Anthropic calls "the most psychologically settled model we have trained to date" — a framing that blurs the line between capability tuning and something that looks a lot like therapeutic intervention. Whether that language is meaningful or marketing, it signals Anthropic's bet that model stability and identity coherence are legitimate safety levers.

Consumer AI & App Ecosystem

Meta AI's standalone app surged from No. 57 to No. 5 on the App Store following the launch of its Muse Spark model, per TechCrunch. The rapid climb suggests Meta's consumer distribution muscle is finally activating around its AI products in a meaningful way.

Google Gemini can now generate interactive 3D models and simulations in direct response to user queries. The Verge reports that users can rotate and interact with AI-generated 3D objects inline — a capability that could significantly differentiate Gemini for STEM education, product design, and scientific visualization use cases.

Claude Platform & Pricing

Anthropic is rolling out an advisor/executor architecture for the Claude Platform, allowing developers to pair Opus as a high-intelligence advisor with Sonnet or Haiku as the executor agent. The pattern lets Opus weigh in on hard decisions mid-task without running every token through the most expensive model — a meaningful cost optimization for complex agentic pipelines. The feature is designed to deliver near-Opus intelligence at a fraction of the per-token cost.

Claude Enterprise pricing is drawing scrutiny from larger teams, with a detailed community thread from an 800-user organization raising concerns about per-seat economics that feel punishing at scale. Separately, iOS subscribers are being warned that purchasing the Max plan through Apple's in-app purchase system costs $50 more than buying directly on the web — entirely attributable to Apple's platform fees.

Claude Code Developer Corner

v2.1.98 — What's New

The v2.1.98 release (full changelog) ships several high-impact features for enterprise and power users:

Google Vertex AI Setup Wizard — An interactive onboarding flow is now accessible directly from the login screen when selecting "3rd-party platform." It walks you through GCP authentication, project/region selection, credential verification, and model pinning. Previously, Vertex AI setup required manual configuration; this brings it to parity with the first-run experience for direct Anthropic API users.

CLAUDE_CODE_PERFORCE_MODE env var — When set, Edit/Write/NotebookEdit operations will fail with a p4 edit hint instead of silently overwriting read-only files. This is a significant workflow improvement for teams on Perforce-managed codebases, preventing silent file clobbers and enforcing proper checkout discipline. If you work in a Perforce environment, this is a must-set.

Monitor Tool for Background Scripts — A new Monitor tool enables streaming events from background scripts, opening the door to long-running agent tasks that emit observable state without blocking the main loop.

Subprocess Sandboxing with PID Namespacing — Subprocess execution now includes PID namespace isolation, tightening the security boundary around agent-spawned processes. Relevant for anyone running Claude Code in multi-tenant or security-sensitive environments.

Security PSA: Auto-Executing npm Code at Startup

A community warning is circulating about the popular ccstatusline setup: if you followed the README and configured your terminal with npx -y ccstatusline@latest, you are automatically downloading and executing the latest version of that package on every new session — without any review or pinning. The recommendation is to pin to a specific version and audit before updating. If you're running this in an automated or CI environment, treat this as a breaking security concern.

In the Wild

A developer demonstrated Claude Code autonomously testing an iOS app — pointing it at a simulator for Wanderlist (a place-curation app) and letting it run. It surfaced real, reproducible bugs within 8 minutes. For teams doing pre-release QA on mobile apps, this is a compelling signal that Claude Code's agentic testing capabilities are maturing beyond toy demos.

Worth Watching

• AI-generated CSAM conviction: An Ohio man has become the first person convicted under a new AI-specific statute for generating sexually explicit images — a legal precedent that will ripple across content moderation and AI liability frameworks.
• Pentagon AI conflict of interest: A US defense official overseeing AI contracts reportedly reaped millions selling xAI stock, raising serious questions about government AI procurement integrity.
• Maine data center ban: Maine is reportedly on the verge of becoming the first US state to ban major new data centers — a policy move that, if it spreads, could reshape where AI infrastructure gets built in the US.
• Mercor breach fallout: The $10B AI hiring startup is dealing with lawsuits and customer exodus following a data breach — a cautionary tale for AI-native companies handling sensitive hiring and credential data.
• Trivy supply chain attack: A detailed post-mortem on a supply chain attack targeting Trivy shows how attackers harvested credentials from secrets managers — highly relevant for any team using Trivy in their CI pipeline.
• Zeteo distributed research community: A Discord-native research collective is attempting to produce peer-quality AI research through structured community collaboration — an interesting experiment in decentralized scientific production.

Sources

• ChatGPT finally offers $100/month Pro plan — https://techcrunch.com/2026/04/09/chatgpt-pro-plan-100-month-codex/
• ChatGPT has a new $100 per month Pro subscription — https://www.theverge.com/ai-artificial-intelligence/909599/chatgpt-pro-subscription-new
• Florida AG announces investigation into OpenAI over shooting that allegedly involved ChatGPT — https://techcrunch.com/2026/04/09/florida-ag-investigation-openai-chatgpt-shooting/
• Florida launches investigation into OpenAI — https://www.theverge.com/policy/909557/openai-florida-investigation
• OpenAI Pauses Stargate UK Data Center Effort Citing Energy Costs — https://www.bloomberg.com/news/articles/2026-04-09/openai-pauses-stargate-uk-data-center-effort-citing-energy-costs
• Is Anthropic limiting the release of Mythos to protect the internet — or Anthropic? — https://techcrunch.com/2026/04/09/is-anthropic-limiting-the-release-of-mythos-to-protect-the-internet-or-anthropic/
• AI on the couch: Anthropic gives Claude 20 hours of psychiatry — https://arstechnica.com/ai/2026/04/why-anthropic-sent-its-claude-ai-to-an-actual-psychiatrist/
• Meta AI app climbs to No. 5 on the App Store after Muse Spark launch — https://techcrunch.com/2026/04/09/meta-ai-app-climbs-to-no-5-on-the-app-store-after-muse-spark-launch/
• Google's Gemini AI can answer your questions with 3D models and simulations — https://www.theverge.com/tech/909391/google-gemini-ai-3d-models-simulations
• We're bringing the advisor strategy to the Claude Platform — https://www.reddit.com/gallery/1sgyt6s
• Pair Opus as an advisor with Sonnet or Haiku as an executor — https://i.redd.it/272ioxmkg7ug1.png
• Claude Enterprise pricing - am I missing something, or are we literally being penalized for scaling? — https://reddit.com/r/ClaudeAI/comments/1sh2scb/claude_enterprise_pricing_am_i_missing_something/
• Never Buy The Subscription In App — https://www.reddit.com/gallery/1sgwa8g
• [claude-code] v2.1.98 — https://github.com/anthropics/claude-code/releases/tag/v2.1.98
• [claude-code] Changelog v2.1.98 — https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md#2198
• PSA: The recommended Claude Code status line command silently auto-executes new npm code every session — https://reddit.com/r/ClaudeAI/comments/1sh0x1r/psa_the_recommended_claude_code_status_line/
• I let Claude Code autonomously test my iOS app. It found real bugs in 8 minutes — https://v.redd.it/38x5vjvqw8ug1
• Ohio man becomes first to be convicted under new AI statute for sexually explicit images — https://www.theguardian.com/us-news/2026/apr/08/ohio-man-convicted-ai-sexually-explicit-images
• US defense official overseeing AI reaped millions selling xAI stock — https://www.theguardian.com/us-news/2026/apr/09/pentagon-ai-xai-emil-michael
• Maine is about to become the first state to ban major new data centers — https://www.gadgetreview.com/maine-is-about-to-become-the-first-state-to-ban-major-new-data-centers
• After data breach, $10B-valued startup Mercor is having a month — https://techcrunch.com/2026/04/09/after-data-breach-10b-valued-startup-mercor-is-having-a-month/
• How the Trivy supply chain attack harvested credentials from secrets managers — https://vaultproof.dev/blog/trivy-supply-chain-attack
• Can strangers in a discord server produce SOTA AI research? Let's find out. — https://reddit.com/r/MachineLearning/comments/1sh1sa5/can_strangers_in_a_discord_server_produce_sota_ai/