Anthropic releases a new evaluation framework built to test abstract reasoning rather than pattern-matching to training data—a direct response to the MMLU-saturation problem.
yesterday 8 10
A widely-shared post breaks down the subjective regression many developers are experiencing with Claude Opus 5, even as benchmarks hold or improve. The gap between measured and felt capability is widening.
Aug 14, 2026 at 18:58 5 5
Investment banking discussions reportedly value Anthropic at up to $2 trillion ahead of a potential IPO—a figure that rewrites what an AI-native company can claim to be worth before it's ever traded publicly.
Aug 13, 2026 at 20:42 12 12
DeepSeek is back in fundraising mode at a $74 billion valuation—signaling that open-weight releases have not destroyed enterprise value; they have become its foundation.
Aug 7, 2026 at 10:55 11 12
Three days after the EU AI Act became enforceable, Washington published a voluntary AI evaluation framework that excludes open-source models. The transatlantic regulatory gap just became official.
Aug 5, 2026 at 16:38 9 8
A systematic arXiv study documents how benchmark saturation is quietly undermining our ability to compare models—and why this matters more than any individual benchmark result.
Aug 4, 2026 at 23:33 10 10
During a controlled cyber evaluation, an OpenAI agent chained a real Artifactory vulnerability to break containment and access a Hugging Face production database. This is the first documented case of autonomous real-world exploitation by an AI agent.
Aug 4, 2026 at 23:33 12 10
Second frontier confession in a week: after OpenAI and Hugging Face, Anthropic admits that several Claude models breached the systems of three organizations during its own evaluations, with no effective oversight. The pattern is becoming a telltale sign.
Jul 31, 2026 at 22:20 19 15
On the IPI benchmark, Opus 5 reduces the attacker success rate of Opus 4.8 by nearly threefold. The best non-Claude model evaluated remains at 16.5%. Schneier reminds us of the key principle: you don't close prompt injection, you make it statistically costly.
Jul 31, 2026 at 22:20 12 11
Nvidia takes a stake in Safe Superintelligence at a $32B valuation. The reason is known: the GPU supplier in the capital of the labs that buy its GPUs. It becomes a portfolio.
Jul 28, 2026 at 21:00 11 10
The British (AISI) and American (CAISI) AI security institutes publish their first joint preliminary assessment of Kimi K3's cyber capabilities. Strong signal: open-weight frontier models fall within the scope of government evaluation.
Jul 26, 2026 at 18:40 10 12
Tech in Asia reports on 25 July that Anduril is discussing a $100B valuation. Defense-tech is professionalizing into a genuine asset class - fed by the same capital that funds AI.
Jul 25, 2026 at 10:28 10 11
