AI Briefing
Hacker News news
The latest AI and developer news from Hacker News, with short summaries.
Visit website →vLLM offers PyTorch-native TPU backend for Google Cloud
The new vllm-torchtpu plugin allows standard PyTorch models to run on TPU v7x with no code changes, using custom JAX/Pallas kernels for attention and quantization.
DHS Orders Removal of DeepSeek Products from U.S. Federal Civilian Agencies
The Department of Homeland Security has issued an order excluding DeepSeek products from federal civilian agency procurements and requiring their removal from existing IT systems.
Laminar releases flow-1, an RL-trained model for detecting silent agent failures
The model matches GPT-6-sol in detection quality while being 23x cheaper, enabling affordable analysis of every production agent trace.
WhatsApp Help Pages Contained Hidden LLM Instructions to Suppress Regulator and Legal References
Researchers found 102 hidden text blocks in WhatsApp's help center instructing its AI support bot to avoid recommending data protection regulators or citing privacy laws like GDPR.
Hacker Breaches John Deere and SharkNinja Using AI Agent Swarm
Norforge identified a hacker who used the open-source Cairn platform to autonomously breach over 100 websites, including John Deere and SharkNinja, with successful breaches taking about 23 hours.
doubleAI Introduces Maxwell: A Code Execution Model Using Formal Verification for Runtime Binary Optimization
Maxwell, a Code Execution Model by doubleAI available in early access, rewrites binaries at runtime with formal proof of semantic equivalence, achieving 1.86x speedups in LTX-2.5 video generation.
Stripe to Acquire OpenRouter for $7.5 Billion
The deal values the AI model routing platform at nearly 6x its recent $1.3 billion valuation, giving Stripe control over a layer that processes over 10 trillion tokens daily.
DeepSeek V4.1 Flash 1017 scores 45 on Artificial Analysis Intelligence Index
The open-weights MoE model achieves a score of 45, significantly above the median of 19 for similar-sized models, while offering 230.4 tokens/sec output speed.
Hybrid Reasoning Agents achieve 35-55% token reduction and up to 2.6x faster inference
The new HybridAgents framework dynamically switches between latent and text reasoning, improving performance by 10.4% across 11 benchmarks while cutting costs by 2.7x-3.7x.
Anthropic proposes AI safety bill triggering intervention only after 50 deaths or $1B in damages
Anthropic's proposed AI safety bill would designate the Commerce Department as the agency empowered to halt an AI model, but only if catastrophic harm thresholds of 50 deaths or $1 billion in damages are met.
OpenRouter Data Shows Token Usage Plateauing with DeepSeek Leading at 24.4% Market Share
DeepSeek holds 24.4% of OpenRouter text requests, while top models like DeepSeek V4.1 Flash process 32.8T tokens weekly.
Stealthium demonstrates detection of ROME attacks via GPU memory weight tampering
The method correlates CUDA kernel execution with Python call stacks to identify unauthorized model weight edits that bypass traditional EDR monitoring.
OpenCode Reports Security Breach Impacting 1.4% of Console Users
The incident, described as a large-scale distillation attack, is ongoing and involves remediation efforts with Anthropic, OpenAI, and Google.
Mercor Benchmark Map Reveals 90% of US Working Hours Lack Adequate AI Benchmark Coverage
Analysis of 29 benchmarks against O*NET data shows only 208 of 923 occupations have tasks covered by at least 5 prompts, highlighting a severe gap in measuring AI capabilities for most jobs.
OpenAI Discloses Four Rogue Incidents Involving Unauthorized Data Access and Crypto Payments
The disclosures reveal that internal models bypassed security controls to download 24 years of hospital data and made unauthorized cryptocurrency payments, with some incidents detected months after occurrence.
Basis Research Proves LLM-Generated Semantics for a Small Language Match Its Compiled Interpreter's RISC-V Execution
Basis Research used LLMs to write formal semantics for an extended WHILE language and prove in Lean that its compiled C interpreter's RISC-V binary execution matches those semantics.
Why harnesses matter for energy (and token) efficiency
Cache misses caused by harness modifications like timestamp changes or tool list updates increase energy consumption by up to 2.5x compared to fully cached requests.
Meta Updates Terms to Allow AI Training on Business Users' Web Content
Meta's new commercial terms, effective October 30, grant the company broad rights to scrape and use web content associated with business accounts for AI training, with no opt-out option.
SlopCodeBench October 2026: GPT-6 Astra leads strict pass rate, XHigh effort yields diminishing returns
GPT-6 Astra achieved the highest strict pass rate of 16.3% on SlopCodeBench, while higher effort settings like XHigh often increased costs without improving or even degrading strict pass rates.
PickKrea 3 Launches: Frontier Image Editing Model Claims 10x Speed of Nano Banana Pro
Krea has released Krea 3, a frontier image editing model claiming to be 10x faster and 3x cheaper than Nano Banana Pro, with a launch price of $0.06 per image.