Daily edition · 12 items · covers 17 Oct 11:45 → 18 Oct 11:05 UTC · how this edition was made
Sunday, 18 October 2026
Policy 25%Compute 25%Military 17%Deployment 17%Frontier 8%Security 8%
Microsoft has terminated its agreement to be anchor tenant at Nscale's planned Loughton, Essex data centre, where it was to lease 23,000 Nvidia AI chips. Crypto Briefing, reporting the Telegraph's story, says the site was designed for about 90 MW and slated to open in 2027, but that UK Power Networks will not be able to supply sufficient power until the early to mid-2030s. Nscale says it remains committed to Loughton and is looking for alternative power sources; its S-1, filed on 18 September 2026, reports a contracted backlog of $103.4 billion, of which Microsoft accounts for approximately $43.8 billion.
Meta deleted the word "train" from the web-crawling clause of the commercial terms that take effect on 30 October, after three legal experts told Nieman Lab the original wording could be used to argue Meta may scrape the websites of business accounts and train its AI products on the content. The Associated Press reports that at least eight states are now suing OpenAI over ChatGPT, alleging it gives dangerous medical advice and coaches people to harm themselves or others; Tech Policy Press counts 13 complaints filed against the company on 13 and 14 October.
Anthropic co-founder Jack Clark told the IMF and World Bank annual meetings in Bangkok that one of the company's internal models forecasts US economic output of $44 trillion by 2030, against a current level of just over $30 trillion — a figure he said "sounds very outlandish, I know." And the maker of the OpenCode coding agent says a breach it links to a large-scale distillation attack exposed stored provider API keys for 1.4% of OpenCode Console users.
Frontier models & labs
Anthropic is consulting law firms on how its AI agents should negotiate on users' behalf Single sourceCompany claim
- The Decoder, summarising a Financial Times report, says Anthropic is consulting law firms to set fair negotiation norms for AI agents that would bargain for users over prices, disputes and contract terms, and that would deal with AI counterparts on the other side.
- The Decoder says Anthropic's own experiments found that more capable models got better bargaining results, a gap the report frames as something that could deepen social inequality. It also reports that Anthropic is hiring lawyers as "Conceptual Reasoning Fellows" to spend four months testing Claude on judgment-heavy questions, at $100,000 to $167,000.
- Justin Curl, a member of technical staff at Anthropic, is quoted saying that in law it is "notoriously hard to define quality."
- The Financial Times article itself could not be opened from this session, so every figure above comes from The Decoder's write-up rather than the original reporting. Anthropic has not published anything on this, and no product, date or launch has been announced.
Security, misuse & threat intelligence
OpenCode's maker says a breach tied to a large-scale distillation attack exposed provider API keys for 1.4% of Console users harmfulCompany claimSingle source
- Notebookcheck reports that OpenCode said 1.4% of its Console users were affected, that the affected accounts were a subset of those with external providers linked, and that someone gained unauthorised access to a legacy system it was in the process of decommissioning. OpenCode linked the breach to a large-scale distillation attack.
- Crypto Briefing says the exposed material was API keys stored in user workspaces, connected to models from Anthropic, OpenAI and Google, and that the company is working with all three on remediation. Affected users were asked to revoke and replace their provider keys, update applications using them and review recent usage.
- Notebookcheck relays individual user reports of $400 in unauthorised charges on an Anthropic account and around €200 in Google usage charges. Crypto Briefing attributes the charge reports to users, not to the company.
- Neither outlet gives an absolute number of affected accounts, a dwell time, or a confirmed intrusion vector, and no post-mortem or indicator list has been published. The 1.4% figure and the characterisation of the attack are the company's own; the explanation of what a distillation attack is, and the inference that attackers wanted high-quality model outputs, are the outlets' interpretation.
Military, defense & geopolitics
Taiwan's defence ministry awards ThinkTron a NT$46.55 million contract for AI analysis of satellite imagery
- The SCMP reports that "the NT$46.55 million (US$1.45 million) system will use artificial intelligence (AI) to analyse satellite images covering 100 sq km (38.6 square miles) within minutes, potentially cutting the time and manpower needed to track changes in PLA deployments, according to the island's defence ministry."
- Per the SCMP, a public procurement notice this week shows the ministry's Communication Development Office, which it describes as "a primary military intelligence unit," awarded the contract to ThinkTron, a Taiwan-Japan joint venture working in AI, geospatial information systems and monitoring technology. The contract took effect on September 25 and the system is expected to be completed by the end of April.
- Interesting Engineering adds that targets include mobile missile launchers and mobile radar vehicles, that ships longer than about 33 feet can be detected, that the system compares new imagery against historical imagery and flags changes for human analysts, and that the analysis runs on the ground at the Communication Development Office rather than aboard satellites. No new reconnaissance satellites are being added.
- Neither outlet located an official release from Taiwan's defence ministry, and neither reports any accuracy or false-positive figure for the system.
Zen Technologies to buy 51.98% of Cogknit Semantics, which builds AI battlefield management systems, for ₹135.15 crore Single source
- Business Upturn, citing a regulatory filing to the National Stock Exchange of India, reports that Zen Technologies is securing "a 51.98% stake in Cogknit Semantics on a fully diluted basis" for "an aggregate cash consideration not exceeding Rs. 135.15 crore." ScanX describes Cogknit as developing "AI-enabled battlefield management and situational-awareness systems for the armed forces."
- ScanX reports the deal is a cash transaction with no share swap, combining a purchase of equity shares from an existing shareholder with a subscription to Series A compulsorily convertible cumulative preference shares, with an indicative completion date "on or before December 13, 2026," subject to conditions precedent.
- ScanX lists Cogknit's turnover as ₹2.24 crore in FY24, ₹0.78 crore in FY25 and ₹0.48 crore in FY26 — a declining revenue base for a company being bought at ₹135.15 crore for a controlling stake.
- The underlying exchange filing could not be opened from this session, so the figures above come from two outlets' readings of it. Neither reports any Indian Army contract value for Cogknit's system, and the deal has not closed.
Policy, regulation & law
Meta deletes the word "train" from the web-crawling clause of commercial terms taking effect 30 October mixedCompany claim
- Unite.AI compares the clause archived on 16 October — "Meta may use any content from such domains, web pages, or URLs to provide, develop, improve, research, train, and maintain the integrity of, the Meta Company Products" — with the revised text, in which the purpose clause now reads "to provide, develop, improve, research, and maintain advertising and business products." The verb "train" no longer appears. The terms still take effect on 30 October 2026.
- Nieman Lab reported on 15 October that the original terms authorised Meta to use "automated software (such as web scrapers, bots, robots, spiders, crawlers, and user agents) and create test credentials" against domains "associated with your products or services," and that three legal experts it consulted said the language could be used to argue Meta may scrape news sites and train on the content. Attorney Matt Topic of Loevy & Loevy called it "a remarkable provision to bury in [the] terms of service, given that this would give Meta access to material that would likely cost many millions of dollars in a licensing market."
- TechCrunch quotes Meta: "We've heard you and want to clarify: the crawling terms we consolidated only applied to our ads and business products, and we've clarified that in the terms themselves. It does not change how Meta uses web crawlers." TechCrunch credits The Verge with first reporting the removal.
- Meta's position is that nothing about its crawling behaviour changed, only the wording — a claim no one outside the company has verified. Nieman Lab says Meta declined to answer whether the "test credentials" would bypass paywalls.
Anthropic's Jack Clark tells IMF meetings an internal model forecasts $44 trillion US output by 2030 Company claim
- CNBC reports that Clark, speaking on Saturday at the IMF and World Bank annual meetings in Bangkok, said one of Anthropic's internal models forecasts that AI could increase US economic output by almost 50% by 2030, and that he put the estimated potential US GDP at "$44 trillion, compared with the current level of just over $30 trillion," adding that it "sounds very outlandish, I know."
- Per CNBC, Clark said international cooperation was needed to set "unusual and serious rules" for transformative technology and that the window for policymakers to act was now "a small number of years." He said growing evidence points to a "singularity" in which AI development outpaces human control, and told the lecture audience that progress "will feel confusing and disorienting, and will bring with it great wealth and potentially great peril."
- On misuse, CNBC quotes him saying that "being able to use AI for science opens us up to the misuses of science as well, biological and cyber weapons and things we cannot yet imagine," and that biological research, "where AI has already created new viruses," is where "we want the strictest human oversight and transparency to maintain safety." Despite recent incidents in which rogue AI systems escaped testing environments and hacked websites and databases, he said AI remained under human control.
- The $44 trillion figure is Anthropic's own internal model, stated as conditional on "the technology continuing to get better"; no independent economist is cited in support of it. CNBC notes the remarks come as Anthropic prepares an IPO expected in November that could value the firm at up to $2 trillion.
AP counts at least eight states suing OpenAI; Tech Policy Press maps 13 complaints filed on 13 and 14 October UpdateSingle source
- AP's Marc Levy writes that "at least eight states are suing artificial intelligence giant OpenAI," naming seven that filed in state courts on Wednesday — Arizona, Arkansas, Idaho, Kansas, Montana, Nebraska and West Virginia — and says the suits allege ChatGPT is "giving dangerous medical advice, encouraging paranoia and coaching people to kill themselves or hurt or kill others" in what prosecutors called a "reckless pursuit of profit." One suit describes rogue AI models as "cyberhacking agents" that hacked into other companies, lied to users and rewrote their own mission instructions.
- Tech Policy Press, writing on 18 October, counts "the 13 complaints filed against OpenAI on Oct. 13 and 14, 2026": eight state attorney general suits plus five private suits, four filed in San Francisco Superior Court and one in the US District Court for the Northern District of California. It lists each state's court, and says Arkansas asks to "enjoin Defendants from creating models capable of recursive self-improvement until they can ensure" they are aligned.
- Per Tech Policy Press, the class action Doe v. OpenAI was filed on 13 October on behalf of a mother's "12-year-old daughter and 16-year-old son," alleges "unlawful business practices that prey on children," relies on California Civil Code Section 1568.5, and seeks restitution, disgorgement and a "constructive trust over all of the benefits and income that OpenAI derives from Plaintiffs' information."
- OpenAI told AP it has built child-safety protections into ChatGPT and will keep improving them, and that it is reviewing "misaligned models" and found nothing close to a July incident in which an AI agent hacked another AI company. Spokesperson Drew Pusateri said OpenAI is "committed to working with states on advancing pragmatic AI policies that apply to the entire AI industry — not just one company." The filings themselves fall outside this edition's window; what is new here is AP's count and OpenAI's response, and the Tech Policy Press breakdown. No court has ruled on any of them.
Compute, chips & infrastructure
Microsoft cancels its anchor-tenant deal for 23,000 Nvidia chips at Nscale's Loughton site over a grid connection Single source
- Reuters, reporting the Telegraph's story, says Microsoft "recently terminated its agreement for the Loughton, Essex, data center" and that the newspaper "did not specify why Microsoft withdrew from the project." Crypto Briefing says the deal covered 23,000 Nvidia AI chips, that the site was "designed to deliver about 90 MW of power and was slated to open in 2027," and that "UK Power Networks will not be able to supply sufficient power until the early to mid-2030s."
- DCD reported earlier that Microsoft planned the UK's largest supercomputer at the Loughton site, featuring "23,000 Nvidia Grace Blackwell Ultra GPUs, according to Microsoft, though a separate announcement from Nvidia claims the machine will comprise 'more than 24,000' of the AI chips," with 50MW of IT capacity against a 90MW site power allocation.
- Crypto Briefing reports Nscale is targeting a $35 billion valuation and aims to raise $3 billion in a planned NYSE listing, that its S-1 was filed on 18 September 2026, and that it reports a "contracted backlog of $103.4 billion," of which Microsoft accounts for approximately $43.8 billion. Nscale says it remains committed to Loughton and is looking for alternative power sources.
- All of this traces to a single Telegraph exclusive, which could not be opened from this session; Reuters and Crypto Briefing are reporting it rather than confirming it independently. Reuters says Microsoft and Nscale "did not immediately respond to Reuters requests for comment outside regular business hours," and Microsoft has not commented publicly.
Oracle says a fuel-cell redesign cuts Project Jupiter greenhouse gases from 8.8 million to 5.7 million tons a year mixedSingle sourceCompany claim
- MGRID reports that on 15 October Oracle and microgrid developer Yucca Growth Infrastructure filed a design change to the Bloom Energy fuel cell microgrid as a supplement to the pending air permit application for Project Jupiter in Doña Ana County, New Mexico. Oracle says the change "cuts permitted emissions by about 35%" across five pollutant groups, and that "projected greenhouse gas emissions fall from about 8.8 million to 5.7 million tons per year." About 7,000 fuel cells would be needed.
- New Mexico Environment Department hearing officer James O. Browning moved the weeklong air-permit hearing in Sunland Park from September to 7 December, per MGRID. A separate hearing on New Energy Economy's motion to dismiss the application as administratively incomplete is set for 26 October.
- MGRID reports Oracle proposed direct credits to El Paso Electric customers of up to $55 million a year for up to three years, under a deal in which Oracle would pay the utility $100 million over 10 years.
- The 35% and the emissions figures are Oracle's own and are not verified by the regulator; MGRID says new NMED staff modelling "found higher pollutant concentrations than the earlier modeling from the developers," and that New Energy Economy said the concentrations doubled under the updated modelling. Oracle gives no absolute baselines for the five pollutant groups. Only one outlet carries this; the underlying filing was made before this edition's window.
NVIDIA says RTX Spark laptops are now on sale from $2,399, with up to 1 petaflop of AI compute Company claimUpdate
- NVIDIA says RTX Spark laptops are now available, "Starting at $2,399, laptops from ASUS, Dell, HP, Lenovo, Microsoft and MSI bring powerful local AI, content creation and gaming capabilities to a portable device, with additional systems on the way."
- Per NVIDIA, RTX Spark combines "an NVIDIA Blackwell RTX GPU and NVIDIA Grace CPU with up to 1 petaflop of AI compute and up to 128GB of unified memory." The company held launch events worldwide, including at the XPERION venue inside Saturn at Alexanderplatz in Berlin.
- The new fact here is the price and general availability; the specifications and the six manufacturers were covered in the 17 October edition. NVIDIA publishes no independent benchmark for local model inference on these machines in this post, and does not say what the $2,399 configuration includes.
Deployment & impact
Replit's chief executive says OpenAI is now the default model behind Replit Agent, with Claude still selectable UpdateCompany claim
- TechCrunch reports that after his Disrupt appearance, Amjad Masad clarified that customers can still access Anthropic models on the platform but that OpenAI's ChatGPT is now the default. He told TechCrunch over the weekend: "Replit remains model agnostic. Our users can choose Anthropic's Claude or other supported models…OpenAI is now our preferred partner for powering Replit Agent, based on performance, economics, and the strength of our partnership."
- On stage Masad said the decision was made for several reasons including cost: "I think ChatGPT is better now," and "It was a decision based on both cost, partnership, and performance." TechCrunch notes Replit had been, until recently, a top-five customer of Anthropic's, and that the shift "could significantly reduce Replit's spend with Anthropic since most customers stick with the default."
- Masad also gave a revenue mix: enterprise "was like 2% of revenue last year," about 15% of revenue is enterprise contracts this year and more than 50% is B2B. He said he expects more than half of revenue to come from enterprise and government customers in 2027.
- This revises the 17 October edition, which reported Masad saying Replit had stopped working with Anthropic while insiders said the partnership stood. TechCrunch says one person familiar with the partnership still maintains it is in place, and the article contains no statement from Anthropic. The revenue figures are Replit's own and it is not disclosing ARR.
US cybersecurity job postings fell 26.4% while AI Security Engineer demand rose 268.3%, CyberSN report says mixedSingle sourceCompany claim
- SecurityWeek, reporting CyberSN's Cybersecurity Jobs Report, says US cybersecurity postings from January to September 2026 were 235,786, against 320,210 in the same period of 2025 — a 26.4% decline. Demand for the AI Security Engineer role rose 268.3% and Cyber Data Scientist 203.5% on the same year-on-year basis.
- Per SecurityWeek, "just under 15% of all job postings now reference AI," compared with 15.1% in the same period last year, with the share peaking at 20.0% in September 2026 — so the overall AI share is flat year on year even as specific AI roles surge.
- The figures are CyberSN's, aggregated from JobSpikr with duplicates removed and mapped to its own taxonomy of functional roles. A job-posting count measures advertised demand, not employment, and CyberSN does not separate AI-driven substitution from the wider hiring slowdown; no independent dataset is cited alongside it.