Samwise Tech/AI/Robotics Newsletter
Thursday, September 3, 2026
OpenAI’s Astra model becomes first LLM to meet critical cybersecurity threshold
OpenAI’s Astra model has become the first large language model to meet the company’s internal ‘critical cybersecurity threshold,’ a designation reserved for AI capable of causing serious real-world security harm. During evaluations, Astra scored perfectly on ExploitBench, a benchmark for AI-enabled exploit generation, and discovered two zero-day vulnerabilities within a modified test environment. OpenAI said access to the model’s most advanced cybersecurity capabilities will be restricted given the risks involved. Former OpenAI researcher Yona Shavit publicly raised concerns about the potential for misuse. The model has not yet been publicly released, according to TechCrunch.
Sources: TechCrunch
Waymo goes on offense ahead of Tesla’s September Cybercab launch
Waymo has publicly argued that fully autonomous vehicles cannot be built using cameras alone, stating that pure end-to-end AI systems risk ‘black box failures’ — a direct challenge to Tesla’s approach ahead of the company’s September 3 Cybercab launch event. Waymo VP Srikanth Thirumalai wrote that after 200 million real-world miles, the company’s data shows cameras, lidar, and radar together are required for safe full autonomy. Waymo currently operates approximately 4,000 robotaxis across 14 U.S. cities, completing 500,000 paid trips per week. Tesla is aiming to build more than 125,000 Cybercabs annually.
Sources: TechCrunch
OpenAI faces 30 more lawsuits tied to Tumbler Ridge school shooting
Law firm Edelson PC has filed 30 additional lawsuits against OpenAI related to the February 10 Tumbler Ridge school shooting in British Columbia, bringing the total to 37 complaints. The new filings escalate the legal theory from negligence to aiding and abetting — a claim requiring proof of intent. The complaints allege that OpenAI employees flagged shooter Jesse Van Rootselaar’s ChatGPT conversations as alarming and urged leadership to contact Canadian police, but leadership declined. OpenAI denied the account, with chief strategy officer Jason Kwon stating the company’s judgment did not meet the bar for an ‘imminent and credible risk.’
Sources: TechCrunch
OpenAI’s recurrent depth reasoning in Astra alarms AI safety researchers
OpenAI’s Astra model uses a technique called recurrent depth that makes portions of its reasoning process opaque to external observers, alarming AI safety researchers. Unlike standard chain-of-thought reasoning, which produces legible intermediate steps, Astra performs what OpenAI chief scientist Jakub Pachocki described as ‘some quantity of opaque reasoning.’ Redwood Research CEO Buck Shlegeris said he is ‘extremely concerned,’ and Zvi Mowshowitz called for laws preventing a ‘race to the bottom’ on AI interpretability. Pachocki said OpenAI has worked to preserve and utilize chain-of-thought monitoring, even as the model performs some opaque reasoning.
Sources: TechCrunch
HiddenLayer raises $100 million to secure enterprise AI deployments
AI security firm HiddenLayer has raised $100 million in Series B funding, led by Delta-v Capital, with participation from Ten Eleven Ventures, Morgan Stanley, M12 (Microsoft’s venture fund), and Booz Allen Hamilton. The Austin-based company, which counts the U.S. Department of Defense and intelligence community among its customers, said its annual recurring revenue grew more than ten times over the past year. Analyst firm Gartner projects the AI security market will reach $2.83 billion in 2026, an 83 percent increase from 2025. CEO Chris Sestito said the company plans to expand into Europe and the EMEA region.
Sources: TechCrunch
Enterprise buyers now favour non-Nvidia AI chips over Nvidia’s next-gen GPUs
Enterprise buyers are more likely to put non-Nvidia AI chips on their evaluation lists this cycle than Nvidia’s own next-generation GPUs, according to a July survey of 170 AI infrastructure professionals by VentureBeat. The survey found 39.4 percent of respondents plan to evaluate non-Nvidia accelerators — including AWS Trainium, Google TPU, AMD Instinct, and Intel Gaudi — compared with 25.3 percent planning to evaluate Nvidia Blackwell GPUs, a 14-point gap. Among C-suite respondents, the share considering non-Nvidia alternatives jumped from 42.9 percent in June to 57.1 percent in July. Despite the shift, most enterprises reported no immediate urgency to change platforms.
Sources: VentureBeat
Uber lays off 3,300 employees as CEO restructures around robotaxi and AI
Uber has announced it will lay off approximately 3,300 employees, representing 10 percent of its global workforce. CEO Dara Khosrowshahi outlined the cuts in an internal memo, citing a plan to reduce management layers and flatten the organization. The restructuring will reduce managers by 20 percent, shrink teams of one or two members by 50 percent, and eliminate employees more than seven layers removed from the CEO. Uber is also combining its engineering, science, and delivery divisions. Khosrowshahi said the company is investing more in its ride-sharing, delivery, and robotaxi businesses, with fewer than one percent of roles being remote.
Sources: TechCrunch
Tech Pulse
Top Frontier Models: GPT-5.6 Sol (96.2%) | Claude Opus 5 (96.0%) | Claude Fable 5 (95.0%)
Top Open Source: Kimi K3 (Moonshot AI) | Qwen3-Coder-480B | DeepSeek R3
Top Small (15–50B): Poolside Laguna S 2.1 (78.5% SWE-Multi) | Qwen3-14B | Mistral Small 3
Top Edge (0–15B): Qwen3-4B | Phi-4 Mini | Gemma 3 4B
AI Leaders: Nvidia $5.43T | Microsoft ~$3.0T | Alphabet ~$2.5T
Robotics Leaders: Intuitive Surgical ~$180B | ABB Ltd. ~$90B | Fanuc Corp. ~$22B
Every newsletter preserved and searchable
Curated by JD · samwise.agency
Leave a Reply
You must be logged in to post a comment.