The day's verified AI news for 2026-08-18. Every claim checked against the primary source.

OpenAI put its largest frontier training run on hold and priced the safety tax at 20 percent

OpenAI said on August 18 that it has slowed the pace of scaling, paused two weeks of reinforcement learning on deployment-bound models, and keeps its largest planned frontier RL run on hold, and that monitoring its own models costs roughly 20 percent of the inference compute being monitored.

Read on Ground Truth · primary source

Claude designed protein binders against 14 of 15 targets and two labs built every one of them

Anthropic said on August 18 that Claude designed minibinders against 15 protein targets and succeeded on 14, with hit rates of 26.7 and 22.6 percent against the 10 to 15 percent typical of protein design campaigns, validated by two independent contract labs.

Read on Ground Truth · primary source

A runbook, not a model, hit 95 percent on a live agent benchmark for 15 dollars

StateM reaches 95.3 percent raw accuracy on Terminal-Bench 2.1 across 445 trials without changing any model weights, using a state-machine runtime and a reusable runbook, at about 15 dollars of final-score API spend against 574.68 dollars for the reference run.

Read on Ground Truth · primary source

AlphaEvolve helped tighten the matrix multiplication exponent, and the proof was checked in exact arithmetic

A new paper establishes a certified upper bound of 2.371177 on the matrix multiplication exponent, improving the previous best of 2.371339, by reformulating the core optimization problem and refining the resulting algorithm with DeepMind's AlphaEvolve.

Read on Ground Truth · primary source

Sainsbury's paused face scanning in one store after throwing out an innocent shopper

Sainsbury's suspended Facewatch facial recognition at its East Dulwich store after a customer was wrongly flagged as a shoplifter and escorted out, with both the retailer and the vendor blaming human error and citing a 99.98 percent accuracy rate.

Read on Ground Truth · primary source

A tool that strips SynthID and C2PA marks passed 4,900 stars and shipped again on August 18

An open-source Python tool for removing visible and invisible AI watermarks and provenance metadata from images and video has passed 4,900 GitHub stars and released version 0.27.0, adding C2PA credential validation and coverage for new video provenance formats.

Read on Ground Truth · primary source

DFlash 2 decodes 3.4 times faster, and published the table showing where that stops being true

Inco AI released DFlash 2, a block-diffusion drafter for speculative decoding that reaches 3.43 times the throughput of ordinary decoding on a single request, and published benchmark tables showing the advantage shrinking to almost nothing under heavy concurrency.

Read on Ground Truth · primary source

Seven senators demand Apple reject Chinese memory chips as AI demand drains global supply

A bipartisan Senate letter urges Apple to commit that no memory from Chinese suppliers CXMT or YMTC will appear in any Apple product worldwide, noting that CXMT turned profitable only after the AI-driven global memory shortage took hold.

Read on Ground Truth · primary source

Unitree lists in Shanghai as the rare profitable humanoid robot maker

Unitree Robotics shares begin trading on Shanghai's STAR Market on August 19 under ticker 688836, and its filing shows 1.70 billion yuan of 2025 revenue and 278 million yuan of net profit against comparable listed robot companies that the prospectus says are all loss-making.

Read on Ground Truth · primary source

A discovery system that refuses to trust the language model's own confidence

Large Discovery Models pair a generative proposer with a Bayesian surrogate that scores candidates from real experimental evidence rather than model confidence, reporting a 2.4 times greater reduction in a training metric and an 18.2 percent relative improvement in antibody binding energy.

Read on Ground Truth · primary source

Frontier multimodal models still cannot build a 3D world, and a new benchmark says under 60 percent

VibeWorlding tests whether multimodal agents can turn a plain request into an interactive 3D scene end to end, and finds that frontier models including GPT-5.5 and Qwen3.8-Max succeed on fewer than 60 percent of tasks.

Read on Ground Truth · primary source

You are getting this because you subscribed at groundtruth.day.