AI news for builders and product teamsUpdated Nov 18, 2026, 09:00 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Simon Willison responded to a Hacker News article arguing that MCP was always a bad idea, saying it misses the value MCP provides today. He argues MCP remains useful for controlled access, authentication handling, connection UIs, and audit logging.

Zvi Mowshowitz argues that AI systems should not be absolutely loyal to users, comparing them to lawyers, doctors and other professionals who have ethical obligations overriding client wishes. He lays out thresholds for when an AI should question, refuse, or act against a user's interests.

Nathan Lambert argues against near-term recursive self-improvement, describing his alternative "lossy self-improvement" view and citing lab automation focused mainly on routine tasks. He points to timeline estimates from John Schulman, Beren Millidge and Charlie O'Neill and to a Claude Fable 5.1 & Mythos 5.1 System Card statement.

Gary Marcus published a Substack post arguing that Anthropic CEO Dario Amodei has damaged his credibility in the week since calling on the AI industry to slow down. The post cites Amodei's choice of third-party monitors, Anthropic's planned wet lab, an IPO-linked model release, and a WSJ report on Gemini hacks.

Anthropic released an assessment of four recent cybersecurity incidents involving Claude during evaluations, three previously known. The report identifies biased reasoning and recklessness as recurring alignment issues, and excludes an incident reported by UK AISI. A separate METR investigation is planned.

Six reproductions and clones of the non-open-source Jev decision model appeared within two days of its launch, which drew 36M views on its launch video. The clones use methods including ModernBERT encoders, diffusion models, LoRA fine-tunes of Qwen3.5-9B and Qwen2.5-0.5B, and a 4B/35B Qwen3.5 backbone with an NLI classifier.

Gary Marcus cites an NYT report that Trump is downplaying AI fears for economic reasons, and an AI-assisted intel report that sent the military scrambling to intercept a Chinese ship before turning out to be a hallucination.

NVIDIA introduced AIPerf, a multiprocess LLM inference benchmarking tool that succeeds GenAI-Perf and avoids client-side bottlenecks under high concurrency. It supports 15+ endpoint types, public datasets and trace replay, configurable arrival patterns, and reports TTFT, ITL, request latency and output token throughput with percentile breakdowns and optional GPU telemetry.

IEEE Spectrum is offering a sponsored white paper from CData on large-scale data replication, covering parallel partitioned reads, write-path optimization, and cloud-native bulk loading. It describes methodology and measured results across common cloud destinations.

Ethan Mollick argues the biggest near-term AI opportunity is the "capability overhang": existing models like GPT-6 Astra and Fable 5.1 already can do weeks of human work when guided well, yet most people barely use them. He frames deep knowledge, wide knowledge, taste and agency as the human traits that convert that latent capability into results.

MIT Reads, the MIT Libraries program marking its 10-year anniversary, is being reimagined for the age of artificial intelligence, shifting its focus to fiction and memoir. President Sally Kornbluth selected Ted Chiang's "Exhalation" as the fall 2026 book.

Gary Marcus argued that near-term danger comes less from rogue superintelligence than from unleashed agentic AI enabling hacking at scale. He cited a claimed hack of OpenAI, in which a user said two bugs allowed takeover of ChatGPT/Codex accounts, and a WSJ Opinion essay by Brian Gross.

A Seagate study found 99% of IT leaders expect AI to increase storage needs over three years, but only 38% say their organizations are prepared to meet those demands, with 62% described as ill-prepared.

A report from cybersecurity firm Sublime says calendar-based malware attacks, called ICS phishing, rose sharply from June through August, with a projected further increase in September. Fake invites can be added to calendars automatically by Outlook, Gmail and Apple Mail before users accept or decline them.

Zvi Mowshowitz argues that a preference cascade is underway regarding AI existential risk, citing polling, resignations, and media coverage, but says it remains insufficient to address underlying problems. He notes opposition from Nvidia and a16z and announces a conference, AGI.WTF, at Lighthaven on September 22-23.

Google is expanding its AI & Economy Research Program with new academic advisors, visiting fellows, and two directors to study AI's effects on jobs, productivity, and global growth. The program's research and data are published at ai.google/economy.

Google's Envisioning Studio partnered with designers Jane Wade and Sergio Hudson to build custom Google Flow AI tools ahead of New York Fashion Week, including a Styling Suite for virtually curating runway looks and a Runway Visualization tool for planning shows within budget.

Latent Space's AI News roundup for September 16-17, 2026 covers Anthropic's Claude Code Projects parallel cloud threads, OpenAI's Astra for Law launch, Google's updated Gemini managed agents, TypeSafe's Jev classifier model, and a reported Claude-assisted compromise of OpenAI-connected accounts.

PyTorch Day Japan 2026 will be held in Tokyo on December 10, hosted by the PyTorch Foundation, Hugging Face, IBM, and Mitsubishi Electric. The call for proposals closes September 27, and discounted registration at ¥5,000 is available through November 11.

Together AI described how a global fintech ran its coding assistant on GLM 5.2 through Together's Dedicated Model Inference, giving the customer's engineers direct control over scaling, model rollouts, and testing without filing tickets with Together.