An OpenAI safety employee has quit and is sounding the alarm The Verge / TechCrunch

David Robinson, who wrote the safety reports that accompanied every major OpenAI model release, resigned this week and published an editorial in The Atlantic calling the company’s culture “broken.” Robinson led policy planning on the Safety Systems team and describes a pattern of safety concerns being deprioritized as the company accelerates releases. The story landed simultaneously on The Verge and TechCrunch, underscoring its significance across the industry.

🔬 Safety & Culture

An OpenAI Safety Employee Has Quit and Is Sounding the Alarm — The Verge

David Robinson’s resignation is the latest in a string of high-profile departures from OpenAI’s safety-focused staff. In his Atlantic piece Robinson says OpenAI’s cultural commitment to safety has eroded under competitive pressure, and he frames himself — with some self-awareness — as “something of a cliché.” The dual TechCrunch and Verge coverage, alongside his own public essay, suggests this resignation is generating more traction than typical departures.

🚀 Frontier Models

OpenAI Releases GPT-6 Sol and Luna — Cuts Token Prices in Half — The New Stack

OpenAI launched GPT-6 Sol and GPT-6 Luna on September 22, nineteen days after the flagship GPT-6 Astra. Sol is priced at $2/$10 per million input/output tokens — half the cost of GPT-5.6 Sol — and Luna comes in at $0.10/$0.50. Both carry a 1.05M-token context window and are available in ChatGPT Work and Codex for paid tiers. The rapid cadence from Astra to Sol/Luna mirrors Anthropic’s tiered release strategy.

Anthropic Releases Claude Opus 5.5 With Lower Prices and Fable-Level Performance — TechCrunch

Also on September 22, Anthropic shipped Claude Opus 5.5 at $4/$20 per million input/output tokens — 20% below Opus 5 and 40% cheaper on typical workloads. The model matches Claude Fable 5.1 on most benchmarks and is 85% less likely than Opus 5 to attempt to bypass safety boundaries. Sonnet 5.5 and Haiku 5.5 are expected in the coming weeks.

🖥️ Developer Tools & Platforms

Gemini Desktop to Get Sweeping “Full Access” Computer Use Permissions — Testing Catalog

Google is testing a new “Full Access” mode for Gemini Desktop that would let the AI read, create, modify, or delete any file on a Mac, send network requests without per-action approval, and control other apps like Mail and Safari. Safety guardrails remain for purchases, account creation, and legal agreements. The capability is currently in trusted testing with no general availability date announced.

All the AI Agents That Can Live in Your Text Messages — TechCrunch — A roundup of notable SMS-based AI agents spanning general assistants, family tools, travel, and work use cases.

Meta Wants Your Next Gadget to Be Muse-Infused — TechCrunch — Meta open-sourced an ESP32 firmware and Linux SDK for building custom Muse-connected hardware, betting on a hardware ecosystem similar to the one Alexa never quite achieved.

DeepSeek Harness v0.2 Ships for macOS, Windows, and Linux — LLM Stats — Desktop app powered by DeepSeek models for coding and work tasks; version 0.2 adds multi-platform support.

🏛️ Policy & Enterprise

Trump Expected to Name Jay Clayton as AI Czar — LLM Stats — The Director of National Intelligence is expected to take on an AI policy oversight role while retaining his DNI position, according to reporting from October 3.

Amazon Responds to Data Center Backlash, Says It No Longer Uses NDAs — TechCrunch — AWS CEO pushed back against community opposition to new data center buildouts, announcing the company has dropped NDA requirements in local negotiations — a response to criticism over secretive siting practices.

Volantis Raises $88M Series A to Tackle AI’s Memory Bottleneck with Optics — LLM Stats — The startup uses optical interconnects to enable chips with far more fast memory than conventional DRAM approaches, targeting the bandwidth wall that limits large-model inference.

🎮 Industry & Applications

Capcom Is Preparing for a “Future Where We Create Games Together With AI” — The Verge

At Capcom Open Conference RE: 2026, programmer Satoshi Ishida detailed how the studio’s RE Engine is evolving to incorporate AI into the development pipeline, even as its upcoming title Pragmata explores AI horror themes. The talk signals that major Japanese studios are moving past public ambivalence toward practical AI tooling adoption.

Splice CEO Kakul Srivastava Thinks AI Emails Are Killing Conversations — The Verge — The music sample platform CEO argues that AI-drafted emails are eroding authentic communication in the creative industry, in a wider interview about Splice’s direction.

Generated by Claude Sonnet 4.6 on 2026-10-03T10:00:00Z