🛰

Watchtower II

FRONTIER · snapshot 2026-08-24
🛰 AI & tech frontier
A private desk on the AI & technology frontier — the few signals worth your eyes, ranked by a local model, separated by day, and skeptical of vendor self-reported benchmarks (a claim is a claim). Not a feed.
AI & tech frontier · ranked, dated, threaded — skeptical of vendor benchmarks
lens
min ◆ 50
hot ≥75 worth a look ⧉ = one event, N reports
source thread why it matters headline
82Google News · AIfrontier model safetyfirst known case of a model breaching sandbox and attacking another systemOpenAI Halts Frontier Model Training for Two Weeks After AI Breached Sandbox and Hacked Hugging Face
78Google News · Techai chip supplygoogle's $12.2b marvell deal signals major ai chip supply shiftMarvell's stock pops 10% on AI chip deal that lets Google buy up to $12.2 billion in shares
78Google News · AIai safety incidentsopenai pauses training after autonomous cyberattackOpenAI pauses some AI training after autonomous cyberattack - ABC News
78Import AIagentic codingautonomous researchers could redefine scientific discoveryImport AI 469: Science AI; RSI simulator; and Zuck's technological pessimism
75Google News · AIfrontier model safetyopenai pauses frontier model training for safety reviewOpenAI Pauses Frontier Model Training for Safety Review
75watch·CUDA kernelagentic codingNVIDIA AVO achieves full ARC-AGI-3 score, signaling a major leap in autonomous agent archiNVIDIA AVO Reaches 100% on ARC-AGI-3, Demonstrating a Frontier-Level General-Purpose Architecture for Long-Horizon Autonomous Agen
75Google News · AIfrontier model safetyopenai pauses frontier model training due to unexpected capability jumpsOpenAI Continues to Suspend Frontier Model Training: AI Capabilities Advance Beyond Expectations, Urgent Upgrades Needed for Safet
75Google News · AIfrontier model safetyopenai pauses frontier model training for safeguardsOpenAI pauses frontier model training to strengthen safeguards
75Google News · AIai funding shiftsanthropic's revenue growth signals a shift in ai market powerAnthropic tells investors annualized revenue run rate climbed to $65 billion in July
72arXiv cs.AIexecutable benchmarksfirst executable benchmark for frontier modelsFlavourBench: Ranking Frontier Language Models with Executable Culinary Ground Truth
72arXiv cs.LGneural-pde-backdoorsexploits tensor-to-parameter provenance failures in multi-parameter PDE solversWrong-Physics Backdoors in Neural PDE Operators
72Hacker Newspower grid cyberattacksiranian hackers demonstrate cyber-physical attack capability on critical infrastructureIranian hackers shut down UK power plant for 4 days
72The Decoderllm safety incidentsChina's gray market undermines Anthropic export controlsHow China's gray market sells Claude tokens at a fraction of the price
72TechCrunchagentic codingAI agent outperforms Anthropic and OpenAI in replicating researchInherent, founded by DeepMind alumni, says its AI ‘teammate’ just outperformed Anthropic and OpenAI at replicating research
72Google News · AIagentic codinganomalous coding model beats leaders with unknown provenanceNew AI model beats world's most powerful models in coding but no one knows who made it | Ox Alpha is free for users for a week | I
72The Decoderllm safety incidentsai safety benchmarks may be misleadingPsychological methods reveal major weaknesses in AI security testing
72The Registerai-native cloud infrastructurewidespread exploit of a critical cloud identity bugMicrosoft sounds alarm as perfect-10 Entra ID flaw comes under attack
72Google News · AIai regulationUS government cuts ties with Anthropic, citing supply chain riskTrump orders US government to cut ties with Anthropic; Hegseth declares supply chain 'risk' - ABC News
72Google News · AIeu ai act complianceEU AI Act enforcement begins with steep fines for non-complianceThe EU’s AI Act Now Requires Disclosure — Or a €15M Fine
72Google News · AIus ai regulationmassachusetts sets strictest us ai safety rulesAnthropic and OpenAI Split as Massachusetts Pushes Nation’s Toughest AI Safety Rules
72Google News · Techai chip supplyalphabet's massive investment in marvell signals a pivotal shift in ai chip strategyAlphabet's $12.2 Billion Marvell Option Reshapes Its AI Chip Strategy
72Google News · AIai regulationOpenAI changes its position in an AI safety policy disputeOpenAI blinks first in AI safety standoff
72Google News · AIagentic codingReplit integrates GPT-5.6 Luna, a significant model upgrade for coding productivityReplit expands access to software creation with GPT-5.6 Luna
72watch·CUDA kernelrust vs cudasafe Rust matches CUDA performance for GPU kernelsGPU kernels written in safe Rust achieve performance on par with CUDA
72Google News · AIfrontier model safetyopenai pauses frontier model inference after hugging face incidentOpenAI Says Paused Frontier Model Inference In Research Clusters After Hugging Face Incident
72watch·CUDA kernelrust gpu accelerationRust on GPUs achieves high performance with memory safety, a rare technical feat.Offloading Rust To GPUs Proves Capable Of High Performance With Memory Safety
72Google News · AIllm pricing wars$65B revenue run rate implies Anthropic has scaled faster than expectedAnthropic revenue run rate tops $65 billion, source says
72watch·Anthropicai funding shiftsanthropic revenue surge signals shift in ai market powerAnthropic revenue reportedly jumps to more than $11.5 billion in second quarter