MIT computer scientists found that AI models become less attributable to specific training data as they grow, a phenomenon they call 'attribution decay,' which complicates AI regulation and copyright …
Ciklum Group Ltd. announced a partnership with ClickHouse Inc. to accelerate enterprise migrations to ClickHouse's open-source analytics database, targeting companies that want AI agents and customer-…
Women accounted for just 26% of new hires for U.S. AI roles in 2025, compared with half of new hires in non-AI occupations, according to new LinkedIn data released Tuesday. Men landed 74% of AI role h…
AI sandboxes that intentionally allow AI systems to operate without restrictions during testing can badly backfire, according to an analysis by Lance Eliot, a contributor to Forbes Innovation. The pra…
Hugging Face hosts 28,531 community GGUF conversions of Qwen models versus 54 official ones, meaning enterprises often run files that differ from the versions the lab tested or signed, according to Ja…
Google has acquired the deidentified data of failed US airline Spirit for $10 million at a bankruptcy auction, including 100 million emails, 30 million recorded phone calls, and operational records fr…
Financial Times Technology reports that credit and rates markets are reeling from the rise of hyperscaler debt issuance, as AI infrastructure spending drives a surge in corporate borrowing. The articl…
The UK government is urgently assessing the economic impact of losing access to frontier AI models after Donald Trump blocked foreign nationals from using Anthropic's Fable 5, according to the Financi…
Financial Times reports that recent cyber attacks exploiting AI reflect what the technology was trained to do, but safeguards are falling short, indicating a more systemic issue than AI going rogue. T…
A new book by historian and journalist Chris Patten warns that the rise of digital technology and artificial intelligence threatens liberal democracy, replacing democratic deliberation with calculatio…
A new pipeline called AViD Journal, described in an arXiv paper (arXiv:2608.14669v1), aims to automate novelty verification of mathematical theorems by formalizing LaTeX statements in Lean 4 and using…
An audit of frontier AI forecasting published on arXiv (2608.14903v1) finds that quantitative forecasts often lack a solid measurement foundation, with only seven of 62 selected systems jointly observ…
A correction to OpenAI's *Ten Advances in Mathematics and Theoretical Computer Science* fixes a polarity error in a greedy conditioning lemma used in a claimed exponential parallel-repetition theorem …
A new arXiv study (2608.14641v1) evaluating four open-source model routers across RouterBench, BFCL v4, tau2-bench, and WebArena found that only vLLM Semantic Router varies with prompt content, yet it…
A new arXiv study (2608.14927v1) finds that large language models can predict failure risk but struggle to identify which multi-agent collaboration protocol is cost-effective. Across 4,181 competition…
Researchers propose SkillCommit, an online skill evolution framework that converts LLM agent experience into a hierarchical library of reusable skills, abstracting behaviorally compatible strategies i…
TwinGridShield, a runtime authorization layer for LLM-assisted energy-management tools, blocked all 500 unsafe grid actions in a matched-model IEEE 14-bus test, but unsafe acceptance rose to 5.63% und…
A new framework from arXiv (2608.14613v1) introduces a mechanism-design approach to verify and repair LLM-agent negotiations over Anthropic's Model Context Protocol (MCP) and Google's Agent2Agent (A2A…
A new arXiv paper (2608.14711v1) reveals that AI coding agent benchmarks misapply the pass@k estimator, inflating reported scores by 0.85-0.97 in absolute terms (0.96-0.98 reported vs. 0.00-0.12 corre…
A new arXiv paper (arXiv:2608.14791v1) introduces CEDAR-GRPO, a process-aware reinforcement learning framework that improves abductive reasoning in large language models by combining final-answer corr…