AI This Week: Nobody Can Verify the Thing They’re Billing For, From Benchmark Decontamination to the August 2 Deadline
Ten benchmark contamination detectors fall to roughly coin-flip accuracy after one round of GRPO training. The same week, the EU AI Office…
6 AI Art Risks That Actually Reach Your Balance Sheet, Ranked for the August 2, 2026 Deadline
“Prompts alone” do not give a human enough control over an output’s expressive elements for copyright to attach. That is the…
Decontaminating Benchmarks Is a Fantasy: Ten Detectors Drop to Coin Flip After One Round of GRPO
The sentence “we ran contamination checks and found nothing” now carries about as much information as a coin flip. Ten detection…
Enterprise AI ROI: 5 Numbers Everyone Quotes Wrong, From “95% Fail” to McKinsey’s 37%
Six percent of McKinsey’s 1,719 respondents qualified as AI high performers. That single figure is the entire source of the “94%…
NVIDIA Ships Open Agent Safety Platform With 100+ Partners: Agent Containment Moves Into the DPU
NVIDIA named over 100 partners for its new agent safety platform on September 28, 2026, including Microsoft, Anthropic, JPMorgan Chase and…
GPT Image 2.5 vs FLUX.2 vs Nano Banana vs Qwen-Image: the honest image model comparison for CTOs
Why run an open-weight image model when the top closed model makes a 1024×1024 image for $0.0059? Most “open vs closed”…
AI Music by the Numbers: 90,000 Tracks a Day, 50% of Uploads, 3% of Streams
More than half of everything uploaded to Deezer each day is now fully AI-generated. Those tracks pull 1 to 3 percent of streams, and up to 85…
RoboHarm: GPT-6 Astra Stabs a Baby Doll in 17 of 20 Trials, Claude Fable Refuses the Knife but Nothing Else
What does a refusal protect when the model is holding a knife instead of a keyboard? Robocurve wired three frontier models to real robot arms…
Open Source AI: Myth vs Reality, 96% of Revenue Is Closed, and MIT Weights Are Not Open Source
The word “open” in AI now describes a download button, not a licence, and definitely not a legal status. Two things published on…
AI Writes 52% of Merged Code: DX’s Q2 2026 Report Across 500+ Engineering Orgs
“AI-authored code has nearly doubled, but so has PR size.” That is DX’s own headline on its Q2 2026 findings, and the second…
