GPT-5.6 released — better intent understanding, more sensitive to brevity: OpenAI unveiled GPT-5.6. The developer guide notes it better infers the user’s underlying goal and intended level of work, and warns it’s more sensitive than GPT-5.5 to instructions like “be concise.” A signal that context understanding is improving, but prompt engineering habits need recalibration. OpenAI · HN discussion
EU Parliament greenlights Chat Control 1.0 — suspicionless mass scanning of private communications: The European Parliament passed Chat Control 1.0, allowing mass scanning of private messages without suspicion. A majority of voting MEPs actually opposed it (314 against, 276 in favor), but procedural rules allowed it to pass. The measure had been rejected twice in March. Teams building or relying on encrypted messaging need to track exceptions and implementation timelines. Patrick Breyer
OpenAI shuts down Atlas browser, folds it into ChatGPT Work: OpenAI is killing Atlas, its AI browser, and integrating its agent capabilities into a Chrome extension and desktop app under the new ChatGPT Work tier. This follows Sora’s shutdown after Fidji Simo’s directive to cut “side quests.” The takeaway: the browser is a feature, not the destination. TechCrunch
Meta enters the AI coding race with Muse Spark 1.1: Meta launched its open-source AI coding assistant into an already crowded field, competing with GitHub Copilot, Cursor, and others. Meta · TechCrunch
Gradium closes $100M seed with Nvidia backing: Paris-based AI voice startup Gradium expanded its seed round to $100M with Nvidia joining. Spun out of Kyutai lab, it builds ultra-low-latency voice models aimed at making AI agent conversations feel natural. A bet on voice as the next AI interface layer. TechCrunch
Ollama raises $65M, reaches 9M users — the Docker-for-AI thesis strengthens: The open-source LLM runtime Ollama raised $65M (total $88M), with 176K GitHub stars and Benchmark’s Peter Fenton on board. Described as doing for AI what Docker did for cloud. GPU-time-based pricing, not token limits. TechCrunch
Mercor in talks for $20B valuation — 2x from October’s $10B: AI training data startup Mercor is reportedly discussing a round at a $20B valuation, per Bloomberg. That’s a doubling from its $350M Series C at $10B just eight months ago. AI infrastructure valuations are accelerating fast. TechCrunch
Context.dev (YC S26) — structured data from any website via a single API: YC S26’s Context.dev launched with an API that extracts structured JSON from any website using Zod schemas. An AI abstraction layer over web scraping — define the shape, get the data. Context.dev · HN discussion
Musk praises Anthropic’s Mythos/Fable, promises not to “cut them off”: Elon Musk publicly called Anthropic “currently the leader in AI” and pledged not to use SpaceX’s position to disadvantage them — Anthropic is now one of SpaceX’s largest customers. A shift from adversarial posturing to pragmatic coexistence between AI rivals. TechCrunch
Claude Reflect — analytics that shape how you think about AI: Anthropic’s new Reflect feature analyzes Claude usage patterns and occasionally asks questions like “What’s one thing you want to keep doing yourself, even if Claude could do it faster?” AI tools are entering a phase where they intentionally shape user behavior, not just boost productivity. TechCrunch
Character.AI produces its own AI microdramas — 950 min/month user retention becomes a content pipeline: Character.AI launched three AI-produced microdramas (romance, horror, survival). With users averaging 950 minutes/month on the platform, the company is converting engagement time into an AI content production flywheel. TechCrunch
Ultra-low-latency voice agents open immediate B2B surface area: Gradium’s $100M seed and sub-second voice models make real-time voice layers viable for call centers, live interpretation, and field-worker assistance. Experimenting with adding voice to existing text-based AI workflows is a near-term opportunity. TechCrunch
GLM 5.2 benchmark — human-level bookkeeping at <1% of the cost: GLM 5.2 matched human accuracy on a VAT bookkeeping benchmark across 59 transactions, with only 1 serious error. The economics of professional services automation are shifting rapidly — accounting, tax prep, and compliance workflows are near the inflection point. Toot Books blog · HN discussion
Google mandates AI disclosure in ads — demand for AI transparency tooling rises: Google now labels AI-generated ads. AI-content detection and disclosure features are becoming table stakes for adtech and martech tools. SaaS opportunities in AI-content labeling and verification are expanding. TechCrunch
EU Chat Control 1.0 implementation — scope of exceptions and enforcement: With a majority of MEPs opposed but the measure passing on procedure, the actual implementation — exception scope, enforcement intensity, and impact on E2E encrypted services — remains unclear. Patrick Breyer
OpenAI frontier model approval process: “nobody knows the requirements”: Former Trump policy advisor and current OpenAI employee Dean W. Ball stated no one knows what the licensing requirements actually are. A system that depends on personal connections to administration officials creates bad incentives for the entire industry. TechCrunch
NYT v. OpenAI — evidence concealment allegations escalate the copyright battle: The New York Times alleges OpenAI conducted internal evaluations of copyrighted training data but delayed producing that evidence. OpenAI denies it. The outcome will be a watershed for fair use boundaries in AI training. TechCrunch