-
OpenAI withholds GPT-6.1 Astra after internal testing finds safety and alignment problems more…
Also inside: AISI evaluation lockdown · MLCommons Jailbreak Benchmark · Datura tool-chain attacks · OpenAI sandbox escapes · ECB AI cyber deadline · Gemini 4 Argon rollout
8 papers & tools · deep-dive · 5 trends · 25 sources
-
Bank of England flags opaque, circular AI debt financing as a financial stability risk more…
Also inside: FERC PJM auction · Amazon Calvert Cliffs nuclear · AI server smuggling case · Australia data-centre standards · DeepSeek Ascend toolchain · HBM DRAM squeeze
4 papers & tools · deep-dive · 5 trends · 35 sources
-
Shared KV-caches silently corrupt answers and leak prompts across tenants in vLLM, SGLang more…
Also inside: KV provenance contract · OpenAI-Baseten marketplace · Inference Auctions paper · QATFactory quantization · Modal $15.75B round · Fireworks FireRouter
6 papers & tools · deep-dive · 5 trends · 12 sources
-
Study finds AI-generated web text now caps pretraining gains past a threshold more…
Also inside: Gemini 4 Argon · GPT-6.1 Sol pricing · Claude Opus 5.5 · Kimi K3 on Bedrock · Salesforce Koa · DrivingBench VLM test
5 papers & tools · deep-dive · 5 trends · 21 sources
-
New agent memory system re-checks permissions before recalling stored data more…
Also inside: OpenAI Dots agent · MCPA certification · Agent benchmark defects · Decisions API · Mastra 1.72.0 · Failure-transparent agents
8 papers & tools · deep-dive · 5 trends · 17 sources
-
IMF finds official statistics systematically undercount generative AI's productivity gains more…
Also inside: VA Abridge IDIQ contract · ILTA legal AI survey · OTel GenAI conventions · ISO 42001 payments · HPE NVIDIA OpenShell · Multi-model governance preprint
5 papers & tools · deep-dive · 5 trends · 18 sources