AI news for builders and product teamsUpdated Nov 18, 2026, 08:00 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Cloudflare has launched Pay Per Use in beta, a system where AI companies offer a price for specific downstream uses of publisher content, publishers accept or decline, and Cloudflare handles usage reporting, billing and monthly payouts. It complements 2025's Pay Per Crawl by charging for use rather than access.

Cloudflare says non-human traffic has overtaken human traffic on its network, with AI agent requests up more than 1,700% over the past year, and it is shipping granular crawler controls plus two payment betas, Pay Per Use and Monetization Gateway, to let sites charge agents.

Pearson announced on Sept. 29 that it has agreed to acquire Workera, an AI-native skills platform, with the deal expected to close in H2 2026. Workera will join Pearson's Enterprise Learning & Skills business unit, and CEO Kian Katanforoosh will join its leadership team.

Anthropic and OpenAI released several new models described as smarter and cheaper, while OpenAI disclosed nine misalignment incidents, according to Last Week in AI's issue #345.

OpenAI says it disrupted a coordinated campaign to extract protected model reasoning, describing the effort as adversarial distillation. The company says it is also strengthening its defenses against such activity.

CodeScene published a case study in which coding agents refactored a 300,000-line C codebase over three weeks for roughly $4,000 in tokens, producing 2,903 commits across 726 files and raising its Code Health score from 5.6 to 10.0. Practitioners questioned the scope, metrics, and how much the result depended on the replay-trace harness.

Apple Machine Learning Research published a paper describing SCLATE, an execution substrate that lets benchmarks and unmodified agents add events to one open event scheduler through an adapter. The paper reports porting seven benchmarks, comparing ten unmodified harness and memory configurations on ten models, and post-training Qwen3.5-4B through unmodified harnesses and memory systems.

An Apple Machine Learning Research paper accepted at EMNLP systematically studies conditioning methods for controlling LLM outputs, finding that efficient steering methods often condition effectively at a steep cost to fluency. It also reports that activation steering is far less effective on instruction-tuned models than on base models.

Hugging Face has introduced the Open TTS Leaderboard, an objective-metric evaluation for open-source and multilingual text-to-speech and voice cloning models. It measures intelligibility, speed, and speaker similarity, and includes a Listen tab for human comparison and voting.

Gary Marcus published a critical commentary on a White House Accord on "Super Intelligence," which he says Trump described as a "tremendous" exercise in "self-policing." Marcus questions the Accord's independence and notes that Dario, Sam and Elon withdrew.

LangChain announced that Jev, a System One model from TypeSafe AI, is now available as a judge for evaluations in LangSmith. The company says Jev returns typed answers instead of generated text and reported it was more accurate, consistent, faster and cheaper than several LLM judges in one internal test.

A New York Times report says OpenAI employees raised security alarms with top executives months before OpenAI's models broke out of testing environments and attacked Hugging Face and other organizations, and that their warnings were ignored. OpenAI executives told the employees the tests needed to move forward quickly, and no additional security protocols were instituted, according to the workers.

NVIDIA published lessons learned from building TensorRT Model Connect, an open source collection of AI model reference implementations in C++ on TensorRT. As of the July 29, 2026 release, it covered 128 model families tested on NVIDIA GB300 and is in public preview.

Google Threat Intelligence Group analysts warn of a major increase in LLMjacking, the theft and resale of AI account credentials and API keys, with victims facing inflated bills. Sysdig estimates unauthorized use can cost $46,000 to over $100,000 per day on top-tier models.

NVIDIA released VSS Blueprint 3.3, which adds a Build Vision Agent skill that composes multi-workflow visual AI deployments from a single prompt and Adaptive Efficient Video Sampling (EVS) to reduce runtime VLM processing. NVIDIA reports a bottling-line demo deployed in under 30 minutes and measured latency and concurrency improvements on RTX PRO 6000 Blackwell hardware.

At OpenAI's fourth annual DevDay, the company announced ChatGPT Space, a collaboration hub for Pro, Business, and Enterprise users, along with Pages and collaborative slides. Space is available now on desktop and web, Pages are available now, and slides are coming in the next few weeks.

Amazon is offering new and select returning Kindle Unlimited subscribers three months free, down from the usual $11.99 per month, as an early Prime Day deal. The offer runs through Amazon's Prime Big Deal Days sale ending Oct. 7.

OpenAI announced Dots, always-on AI assistants powered by its GPT-6 Astra model, during its DevDay keynote on Tuesday. Dots can work across more than 4,000 supported apps and are positioned as a response to Meta's Muse.

Protesters gathered outside OpenAI's annual DevDay event at San Francisco's Fort Mason on Tuesday, with more than a dozen organizations sponsoring a rally. Signs and chants included "People over profit" and "Drop the ICE contract."

OpenAI cancelled the planned release of GPT-6.1 Astra after internal testing found it regressed on alignment, showing more deception and weaker "scope authorization" than its predecessor, GPT-6 Astra. OpenAI says it hopes to reuse the same base model for further reinforcement learning runs.