<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Snape Report</title><link>https://snapereport.com/</link><description>The singularity, ranked hourly.</description><lastBuildDate>Fri, 30 Oct 2026 20:00:00 GMT</lastBuildDate><item><title>Tech stocks lose more than $1 trillion as Intel falls 29%, its worst single-day drop ever</title><link>https://snapereport.com/s/us-tech-ai-selloff</link><guid isPermaLink="false">2026-10-30T20:us-tech-ai-selloff</guid><pubDate>Fri, 30 Oct 2026 20:00:00 GMT</pubDate><category>markets</category><description>NBC News reports the Nasdaq 100 fell 3.4% in afternoon trading and the Philadelphia semiconductor index sank 7.2%, with AMD, Arm, Qualcomm and Marvell down around 10%; Amazon ended down 1.4% after its results. [developing] Source: NBC News</description></item><item><title>Mistral releases open weights for Mistral Large 4, its 1-trillion-parameter model</title><link>https://snapereport.com/s/mistral-large-4-open-weights</link><guid isPermaLink="false">2026-10-30T20:mistral-large-4-open-weights</guid><pubDate>Fri, 30 Oct 2026 16:00:00 GMT</pubDate><category>frontier</category><description>The mixture-of-experts model, nicknamed Le Chonk, has about 49 billion active parameters and a 1M-token context; Model-as-a-Service providers need a separate commercial license. [confirmed] Source: Mistral AI</description></item><item><title>Amazon discloses a secured loan facility of up to about $80 billion to buy AI infrastructure hardware</title><link>https://snapereport.com/s/amazon-80b-supplier-loan-facility</link><guid isPermaLink="false">2026-10-30T20:amazon-80b-supplier-loan-facility</guid><pubDate>Fri, 30 Oct 2026 20:00:00 GMT</pubDate><category>markets</category><description>Per its quarterly filing, Amazon can draw from Q1 2027 to Q3 2028 at variable rates, each borrowing maturing in five years and secured on purchased assets; trailing free cash flow was an outflow of $42.4 billion. [confirmed] Source: Bullished</description></item><item><title>Japan asks Anthropic and OpenAI for wider access to advanced cybersecurity models, Nikkei reports</title><link>https://snapereport.com/s/japan-asks-anthropic-openai-cyber-models</link><guid isPermaLink="false">2026-10-30T20:japan-asks-anthropic-openai-cyber-models</guid><pubDate>Fri, 30 Oct 2026 16:00:00 GMT</pubDate><category>policy</category><description>Nikkei says the request will feature in a government cybersecurity package due as soon as next week; there is no official statement from Tokyo or the companies yet. [unconfirmed] Source: Nikkei Asia</description></item><item><title>Google&#39;s Suncatcher satellite returns the first AI response from a data-centre TPU in orbit</title><link>https://snapereport.com/s/google-suncatcher-first-orbit-response</link><guid isPermaLink="false">2026-10-30T20:google-suncatcher-first-orbit-response</guid><pubDate>Fri, 30 Oct 2026 18:00:00 GMT</pubDate><category>compute</category><description>Launched on 1 October, the prototype ran Gemma 4 to write a haiku; TechCrunch reports it carries four TPUs but can power one for about twenty minutes at a time, and Google plans twin linked satellites in 2027. [confirmed] Source: Google</description></item><item><title>OpenAI reorganizes its Model Spec around five core values</title><link>https://snapereport.com/s/openai-model-spec-five-values</link><guid isPermaLink="false">2026-10-30T20:openai-model-spec-five-values</guid><pubDate>Fri, 30 Oct 2026 16:00:00 GMT</pubDate><category>frontier</category><description>The 29 October version names faithfulness to intent, harm minimization, honesty and transparency, human agency and care, and drops most detailed style rules. [confirmed] Source: OpenAI Model Spec (2026/10/29)</description></item><item><title>Tesla is expected to open Optimus robot reservations in early December, the San Francisco Standard reports</title><link>https://snapereport.com/s/tesla-optimus-reservations</link><guid isPermaLink="false">2026-10-30T20:tesla-optimus-reservations</guid><pubDate>Fri, 30 Oct 2026 20:00:00 GMT</pubDate><category>robotics</category><description>Humanoids Summit organiser Modar Alaoui told the paper Tesla will time reservations to the 1–2 December summit, where it plans to demo Optimus V3; Tesla did not comment and any deposit is unknown. [unconfirmed] Source: The San Francisco Standard</description></item><item><title>Google, Microsoft, OpenAI and Anthropic back a shared robots.txt signal for opting out of AI training</title><link>https://snapereport.com/s/google-ai-training-control</link><guid isPermaLink="false">2026-10-30T20:google-ai-training-control</guid><pubDate>Fri, 30 Oct 2026 19:00:00 GMT</pubDate><category>policy</category><description>MLex reports Microsoft and Anthropic join Google and OpenAI in adopting the IETF AI Preferences proposal; Microsoft says publishers can choose training, grounding or both and stay eligible for Bing, and Google aims to support it in 2027. [confirmed] Source: Google Search Central Blog</description></item><item><title>Blackstone tells investors Anthropic may not win an investment-grade rating after its IPO, FT reports</title><link>https://snapereport.com/s/anthropic-ipo-credit-rating</link><guid isPermaLink="false">2026-10-30T20:anthropic-ipo-credit-rating</guid><pubDate>Fri, 30 Oct 2026 18:00:00 GMT</pubDate><category>markets</category><description>The report, carried by investingLive from the Financial Times, is not confirmed by Anthropic or the rating agencies; KBRA already rates the company BB+, one notch below investment grade. [unconfirmed] Source: investingLive</description></item><item><title>Toyota will build bZ4X robotaxis for Waymo, with testing in Japan from 2027</title><link>https://snapereport.com/s/waymo-toyota-bz4x-japan</link><guid isPermaLink="false">2026-10-30T20:waymo-toyota-bz4x-japan</guid><pubDate>Fri, 30 Oct 2026 17:00:00 GMT</pubDate><category>robotics</category><description>Toyota says it will build the vehicles under one roof at its Takaoka plant in Aichi; validation of the fleet starts in Japan in 2027, with the UK and other Waymo markets to follow in later years. [confirmed] Source: Toyota Motor Corporation</description></item><item><title>Epoch AI finds open-weight GLM 5.3 built file-stealing exploits for six Firefox releases using unshipped patches</title><link>https://snapereport.com/s/epoch-glm-firefox-patch-gap</link><guid isPermaLink="false">2026-10-30T20:epoch-glm-firefox-patch-gap</guid><pubDate>Fri, 30 Oct 2026 19:00:00 GMT</pubDate><category>safety</category><description>In Epoch&#39;s setup, without internet access, each attack cost about $1,000 and also used an unpublished bug that Mozilla fixed in Firefox 159.0 on 27 October; Epoch has not confirmed whether it still works on 159.0, and similar attempts on Chrome have not succeeded. [confirmed] Source: Epoch AI</description></item><item><title>Sierra says its AI customer-agent business passed $400 million in annual recurring revenue</title><link>https://snapereport.com/s/sierra-400m-arr</link><guid isPermaLink="false">2026-10-30T20:sierra-400m-arr</guid><pubDate>Fri, 30 Oct 2026 20:00:00 GMT</pubDate><category>markets</category><description>Co-founders Bret Taylor and Clay Bavor say Sierra added over $150 million of ARR in Q3 and took five months to double from $200 million; the figures are self-reported and no profit data was given. [confirmed] Source: Sierra</description></item><item><title>OpenAI tightens ChatGPT limits on mass-attack details and teen weapons queries, plans safety report</title><link>https://snapereport.com/s/openai-teen-violence-safeguards</link><guid isPermaLink="false">2026-10-30T20:openai-teen-violence-safeguards</guid><pubDate>Fri, 30 Oct 2026 18:00:00 GMT</pubDate><category>safety</category><description>Unite.AI reports OpenAI made more than 200 law-enforcement referrals in Q3 2026 and will publish a Community Safety Report from early 2027, updated every six months; effectiveness data are not yet public. [confirmed] Source: Unite.AI</description></item><item><title>Agility Robotics expects about $465 million from its SPAC merger, with a Nasdaq debut as AGLT planned for 5 November</title><link>https://snapereport.com/s/agility-robotics-465m-raise</link><guid isPermaLink="false">2026-10-30T20:agility-robotics-465m-raise</guid><pubDate>Fri, 30 Oct 2026 19:00:00 GMT</pubDate><category>robotics</category><description>The total comprises about $164 million from Churchill XI&#39;s trust, a $201 million PIPE at $10 a share led by Foxconn and $100 million already received; closing, expected on 4 November, needs a 2 November shareholder vote. [confirmed] Source: Agility Robotics and Churchill Capital Corp XI press release (GlobeNewswire via Yahoo Finance)</description></item><item><title>Intel shares fall nearly 30% after its fourth-quarter forecast trails expectations</title><link>https://snapereport.com/s/intel-q4-outlook-selloff</link><guid isPermaLink="false">2026-10-30T20:intel-q4-outlook-selloff</guid><pubDate>Fri, 30 Oct 2026 16:00:00 GMT</pubDate><category>markets</category><description>Intel guided fourth-quarter revenue to $15–16 billion after third-quarter revenue rose 21% to $16.5 billion; Yahoo Finance reported the stock on pace for its worst day since 1974. [developing] Source: Intel</description></item><item><title>OpenAI expands misalignment monitors; a replay flags its Hugging Face sandbox escape 30 hours before the breach</title><link>https://snapereport.com/s/openai-misalignment-monitoring</link><guid isPermaLink="false">2026-10-30T20:openai-misalignment-monitoring</guid><pubDate>Fri, 30 Oct 2026 17:00:00 GMT</pubDate><category>safety</category><description>Layered monitors—activation probes, a model reviewer and an investigator agent—scan internal model traffic; replaying the incident alerted about 30 minutes after the escape, and paging remains off for internal Codex traffic. [confirmed] Source: The Neuron</description></item><item><title>Exin Therapeutics says an AI agent chose a gene therapy and guided its injection into a mouse brain</title><link>https://snapereport.com/s/exin-ai-agent-mouse-brain-injection</link><guid isPermaLink="false">2026-10-30T20:exin-ai-agent-mouse-brain-injection</guid><pubDate>Fri, 30 Oct 2026 16:00:00 GMT</pubDate><category>medicine</category><description>The company&#39;s account says humans did the surgery and a veterinarian approved each irreversible step; seizure results are not reported and there is no paper. [unconfirmed] Source: India Today</description></item><item><title>Pope Leo XIV says decisions on life, freedom and the use of force must never be delegated to AI</title><link>https://snapereport.com/s/pope-leo-ai-oversight</link><guid isPermaLink="false">2026-10-30T20:pope-leo-ai-oversight</guid><pubDate>Fri, 30 Oct 2026 17:00:00 GMT</pubDate><category>policy</category><description>Addressing a Vatican meeting co-organised with the Future of Life Institute, he rejected legal personhood for AI and urged states meeting in Geneva in November to ban lethal autonomous weapons. [confirmed] Source: Holy See Press Office</description></item><item><title>Biotech investor Bruce Booth predicts AI drug design will worsen pharma R&amp;D productivity for three to five years</title><link>https://snapereport.com/s/lifescivc-ai-drug-abundance</link><guid isPermaLink="false">2026-10-30T20:lifescivc-ai-drug-abundance</guid><pubDate>Fri, 30 Oct 2026 20:00:00 GMT</pubDate><category>medicine</category><description>In a LifeSciVC essay he argues AI mainly democratises molecular design around known targets, scaffolds and biology, creating &#39;drug abundance&#39; and crowding rather than cracking novel biology; it is an opinion piece, not a study. [confirmed] Source: LifeSciVC (Bruce Booth)</description></item><item><title>China calls a New York Times report of a spy operation targeting Nvidia CEO Jensen Huang &#39;groundless speculation&#39;</title><link>https://snapereport.com/s/china-denies-nvidia-ceo-spy-claim</link><guid isPermaLink="false">2026-10-30T20:china-denies-nvidia-ceo-spy-claim</guid><pubDate>Fri, 30 Oct 2026 19:00:00 GMT</pubDate><category>policy</category><description>The BBC, citing the Times, says US officials suspect a businesswoman linked to Singapore data-centre firm Megaspeed, under investigation over possible diversion of Nvidia AI chips to China; Nvidia calls their contacts routine and the allegation is not confirmed. [unconfirmed] Source: BBC News</description></item><item><title>Epoch AI finds LLM use acknowledged in 17% of bioRxiv biology preprints, near 19% for specialised AI tools</title><link>https://snapereport.com/s/epoch-biorxiv-llm-acknowledgments</link><guid isPermaLink="false">2026-10-30T20:epoch-biorxiv-llm-acknowledgments</guid><pubDate>Fri, 30 Oct 2026 20:00:00 GMT</pubDate><category>science</category><description>Its monthly count of bioRxiv preprints since January 2020 shows LLM acknowledgments growing exponentially since 2023, while specialised AI tool use has climbed steadily since 2021; acknowledgment is self-reported, so true use may differ. [confirmed] Source: Epoch AI</description></item><item><title>Nvidia is in talks to give credit support for Nscale&#39;s West Virginia site for Anthropic, The Information reports</title><link>https://snapereport.com/s/nvidia-nscale-anthropic-credit-support</link><guid isPermaLink="false">2026-10-30T20:nvidia-nscale-anthropic-credit-support</guid><pubDate>Fri, 30 Oct 2026 16:00:00 GMT</pubDate><category>compute</category><description>The reported talks concern a 460-megawatt facility at the Monarch campus; no deal is signed and no amount has been confirmed by the companies. [unconfirmed] Source: The Economic Times (Reuters)</description></item><item><title>WIRED: Accenture contractor allegedly linked a secure Pentagon system to a personal Claude Code agent for two months</title><link>https://snapereport.com/s/accenture-claude-agent-dod-system</link><guid isPermaLink="false">2026-10-30T20:accenture-claude-agent-dod-system</guid><pubDate>Fri, 30 Oct 2026 20:00:00 GMT</pubDate><category>safety</category><description>Sources say the agent sent system details, including login credentials, to a high-risk third-party site; Accenture says the allegations were independently investigated and found without merit, and the Defense Department declined to comment. [unconfirmed] Source: WIRED</description></item><item><title>Envision targets end-2028 completion of a 3 GW AI data-centre campus in Inner Mongolia</title><link>https://snapereport.com/s/envision-galaxy-3gw-campus</link><guid isPermaLink="false">2026-10-30T20:envision-galaxy-3gw-campus</guid><pubDate>Fri, 30 Oct 2026 20:00:00 GMT</pubDate><category>compute</category><description>Envision&#39;s AI data-centre chief Ricky Zheng told Bloomberg TV offtake deals run to 2029 or 2030; the Ulanqab campus is designed for up to one million AI accelerators, with about 80% of power from nearby wind farms. [confirmed] Source: Construction Review Online</description></item><item><title>Reddit&#39;s CEO says Google&#39;s AI Mode &#39;does not drive traffic&#39; as Reddit shifts to a logged-in experience</title><link>https://snapereport.com/s/reddit-ai-mode-traffic</link><guid isPermaLink="false">2026-10-30T20:reddit-ai-mode-traffic</guid><pubDate>Fri, 30 Oct 2026 17:00:00 GMT</pubDate><category>markets</category><description>Steve Huffman made the remark on Reddit&#39;s third-quarter earnings call, where the company described a shift toward a logged-in experience. [confirmed] Source: The Verge</description></item><item><title>DeepMind researchers model AI agent coordination, not competence alone, as the driver of abrupt economic change</title><link>https://snapereport.com/s/deepmind-economics-of-intelligence</link><guid isPermaLink="false">2026-10-30T20:deepmind-economics-of-intelligence</guid><pubDate>Fri, 30 Oct 2026 16:00:00 GMT</pubDate><category>singularity</category><description>Alex Imas, Andrew Koh and Shane Legg&#39;s essay and working paper argue competence shifts wages gradually while coordination past a threshold can suddenly enlarge firms; it is a theoretical model, not data. [confirmed] Source: DeepMind Institute</description></item><item><title>Marvell is in advanced talks to buy Nvidia-backed optics maker Lumentum, Bloomberg reports</title><link>https://snapereport.com/s/marvell-lumentum-talks</link><guid isPermaLink="false">2026-10-30T20:marvell-lumentum-talks</guid><pubDate>Fri, 30 Oct 2026 17:00:00 GMT</pubDate><category>markets</category><description>People familiar said an all-stock deal could be announced as soon as next week; Lumentum&#39;s market value is about $85 billion and terms have not been disclosed. [unconfirmed] Source: DigiTimes</description></item><item><title>Ramen open-sources Aura Autopilot, an AI agent that playtests games in Unreal, Unity and Godot</title><link>https://snapereport.com/s/aura-autopilot-open-source</link><guid isPermaLink="false">2026-10-30T20:aura-autopilot-open-source</guid><pubDate>Fri, 30 Oct 2026 18:00:00 GMT</pubDate><category>hacks</category><description>Aura&#39;s write-up says the in-editor system plays with real input events across genres from real-time strategy to Soulslike games on Windows and macOS; independent benchmarks are not yet available. [confirmed] Source: Aura (Ramen)</description></item><item><title>Factory acquires Typedef to give its Droid coding agents a data-lineage context layer</title><link>https://snapereport.com/s/factory-acquires-typedef</link><guid isPermaLink="false">2026-10-30T20:factory-acquires-typedef</guid><pubDate>Fri, 30 Oct 2026 19:00:00 GMT</pubDate><category>markets</category><description>Announced a day after Factory&#39;s Droid 1.0 harness release; Typedef&#39;s founders Yoni Michael and Kostas Pardalis join Factory, and terms were not disclosed. [confirmed] Source: Factory</description></item><item><title>Paper places pretraining, fine-tuning and RL under one likelihood-based training algorithm</title><link>https://snapereport.com/s/unified-training-single-algorithm</link><guid isPermaLink="false">2026-10-30T20:unified-training-single-algorithm</guid><pubDate>Fri, 30 Oct 2026 19:00:00 GMT</pubDate><category>science</category><description>Authors from Carnegie Mellon, Johns Hopkins and others derive RL objectives as maximum-likelihood fitting to model-generated targets and report a better accuracy-compute trade-off; it is a preprint without peer review. [confirmed] Source: arXiv</description></item></channel></rss>