Fetching from the wire…
Agents2026-09-15 · source-backed
CiteShade turns the audit mechanism into the attack surface (arXiv 2609.15660). An attacker controlling a single source gets the system to produce an attacker-chosen wrong answer, attribute it to a trusted source that doesn't support it, and leave the correct evidence retrieved but unused. Wrong-answer rate went from 0.01 to 0.68, with a citation laundering rate of 0.84 under explicit instruction and 0.64 with none. Perplexity filtering and citation-support checking each proved insufficient alone; the authors propose a counterfactual check of which source actually drove generation.
Each link below shares sources, entities, or timing with this story.
On Latent Space July 28, OpenAI core product engineering lead Akshay Nathan said Codex and ChatGPT Work combined reached 10 million users within two weeks of the July 9 launch, with monthly actives up more than 10x since January 2026. The number that should reframe your produc...
D-SCAN (SIGIR 2026) found the standard guardrail returns high confidence on compromised output. Their alternative signal is document-level attention dynamics: during a poisoned generation, attention concentrates on the injected document and entropy collapses, versus dispersed...
Built from production query data, it uses live evidence verification instead of static gold answers, which targets the exact failure of deep-research agents: plausible but unverified synthesis. (Latent Space) If you build or evaluate a research or RAG agent, this is a directly...
The Hrazdan facility opened August 8, scaling to 300 megawatts and 70,000+ NVIDIA Rubin and Blackwell GPUs by end of 2027, built on NVIDIA DSX (40% more GPUs on the same footprint) with Dell PowerEdge, Schneider Electric power and Vertiv cooling. NVIDIA intends to invest, foll...
arXiv 2608.00765 compresses retrieved docs into query-conditioned visual representations, sidestepping the trade-off where hard compression is query-aware but weak and soft compression is strong but needs costly offline encoding. Beats both baselines across varying retrieval d...
The study extracted 130 clean atomic state transitions from 707 real issues in SWE-bench Lite and Verified. Plain RAG scored 0.57-0.59 answer accuracy; an LLM reranker didn't help and added latency, about 18 seconds against 2.1. A (subject, relation, object) supersession memor...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.