Company · News
AI news, worldwide. Read with an operator’s eye.
Three stories from around the world every day, summarised in our own words with a link to the outlet that reported them, and three notes a day from founder Edgar Melo Costa on what they mean for people who build and run businesses. Nothing here is a press release.
Founder’s notes
From Edgar Melo Costa
Opinion pieces by the founder of ELP Ventures. They draw on the day’s reporting and on how we run the company; they are not news reports.
If you would not show it to the customer, do not show it to the model
WhatsApp’s help pages carried hidden notes steering chatbots away from naming regulators. Whatever the cause, it shows why an assistant’s instructions should be ones you could publish.
Thirty billion dollars is a financing, not a business model
The largest AI labs are raising at trillion-dollar valuations while chip stocks swing on debt worries. We chose the other path: price above cost from the first customer and grow on results.
Fifty dollars a breach is the new price of a weak permission
One attacker reportedly used AI agents to break into more than a hundred company sites at about $50 each. The defence is not a better model; it is fewer things an agent can do without a person.
Chinese models are good. Read the terms before you deploy one.
DeepSeek shipped another strong update this weekend. For a business, the model card is the easy part; governing law, data location and training defaults decide whether you can use it.
Regulation is a product specification, if you read it that way
California wants independent verification and kill switches for agents; the UN wants binding rules. We built approval gates and canaries before anyone required them, because customers wanted them.
Electricity is the moat, not the model
Microsoft walking away from a UK site because the grid will not be ready until the 2030s tells you where the real constraint is. Our Ontario plan is about owned capacity for exactly that reason.
World
AI news from around the world
Summaries written by ELP; headlines, facts and quotations belong to the outlets linked. Items are added daily and kept for the record.
DHS directs federal civilian agencies to remove DeepSeek products
Agencies must remove DeepSeek services and third-party tools built on its technology and stop buying them, with limited exceptions, citing data stored on servers in China.
Fox News live coverage ↗OpenAI reports large gains on COBOL migration with GPT-6.1 Sol
On OpenAI’s own, unverified evaluation the model scores up to 97.9%, against 5.4% for its predecessor; the company says production use still needs expert review.
Dataconomy ↗Amazon plans more than €16 billion of investment in Germany in 2027
Announced in Berlin, the plan includes €4.8 billion to expand the AWS Frankfurt region, plus satellite internet, logistics and Prime Video productions.
About Amazon ↗AstraZeneca and Oracle sign an AI pact to speed clinical-trial recruitment
The partnership will use AI and real-world data to find trial participants and monitor safety, starting with a hypertension study in Abu Dhabi with M42.
The Star (Malaysia) ↗ByteDance, Alibaba and Tencent reportedly order over $35 billion of export-compliant Nvidia cards
The orders for more than three million RTX PRO 5500 inference cards may be only partly filled, with some buyers told to expect 5–10% of requested volumes; deliveries are expected from early 2027.
The Edge Malaysia (citing Reuters) ↗DeepSeek open-sources DeepInfer, its production inference engine
Released under Apache 2.0, the framework runs DeepSeek models at scale on Nvidia SM100 GPUs and Huawei Ascend 950 NPUs.
GitHub (deepseek-ai) ↗Chip stocks end their longest losing streak in 18 years as the Nasdaq rises
Semiconductors broke an eight-session slide on Monday, though investors remain focused on the debt financing chips and data centres, with big-tech earnings due midweek.
Fox News live coverage ↗OpenAI reportedly races to raise about $30 billion before Anthropic’s expected listing
The round could value OpenAI above $1.4 trillion, with SoftBank and MGX reported in talks. Anthropic says quarterly revenue grew from $4.8 billion in Q1 to $16.8 billion in Q3.
PYMNTS (reporting Reuters) ↗WhatsApp help pages carried hidden instructions aimed at AI systems
A review found 102 hidden blocks, two of which steered chatbots away from naming data-protection regulators. WhatsApp said they were support-bot instructions exposed by a bug, now fixed.
Redact ↗US sanctions five Chinese entities over alleged theft of Google AI trade secrets
The State Department named Shanghai Zhisuan Technology, founded by a former Google engineer, and four investment entities linked to the MiraclePlus accelerator under the Protecting American Intellectual Property Act.
Export Compliance Daily ↗Meta details Prometheus, a gigawatt-scale AI cluster in Ohio aiming for about a million GPUs
Phase one in New Albany spans hundreds of thousands of GPUs, with racks drawing 135 kW today and over 170 kW for the next generation.
Meta Engineering ↗Researcher: one attacker used AI agents to breach 100+ company websites for about $5,000
A security researcher told Reuters the autonomous agents tried more than 1,200 sites across four countries at roughly $50 of compute per breach. Named companies said impact was contained or under investigation.
Fox News (citing Reuters) ↗Toronto’s Tenstorrent raises $1.5 billion at a $6.2 billion valuation
The Jim Keller-led chip company will scale production of its Galaxy servers and fund RISC-V research; the Series E was co-led by Trace Capital and Innovation X.
SiliconANGLE ↗Documents describe a US Army programme for autonomous offensive cyber agents
The Intercept reports that Army Cyber Command’s ECHOTRIBBLE would build agents that move through networks with minimal human oversight; experts warned of collateral damage.
The Intercept ↗Interpol launches a public campaign against AI deepfake scams
The campaign uses AI-generated demonstration videos and advises people to verify requests independently, agree a family safe word and never send money in response to urgent messages.
Fox News live coverage ↗Anthropic commits 5% of its shares to a new public-benefit fund
Subject to stockholder approval, the fund can receive 2% more for each 50% rise in valuation above a baseline and may not be used for lobbying or political giving. It follows a $200 million commitment to study AI’s effects on workers.
Anthropic ↗Anthropic says its draft federal AI safety bill will change on state preemption
The company wants states to keep authority over child protection and whistleblower rules, and is weighing stronger incident reporting and independent evaluations.
Fox News live coverage ↗Semiconductor index falls for an eighth straight session; memory names lead the decline
The Philadelphia Semiconductor Index slid again on Friday with memory suppliers hit hardest, while a major AI cloud provider fell sharply on debt concerns.
BigGo Finance ↗DeepSeek releases V4.1-Flash-1017, a faster and leaner update aimed at US rivals
The Hangzhou lab published an updated model it says completes tasks faster with shorter outputs and better benchmark scores, continuing China’s cadence of frequent open-weight releases.
Bloomberg via Yahoo Finance ↗Steve Eisman: an OpenAI failure would tip the US into recession
The investor argued that OpenAI’s spending commitments are now large enough that a collapse would immediately hit US growth, while stressing he was not predicting one.
Stocktwits ↗Oracle proposes fuel-cell redesign to cut permitted emissions at its New Mexico data centre
The company says the redesign would reduce the permitted cap by about a third ahead of a December hearing; critics note a permit cap is not the same as measured emissions.
mgrid.org ↗Brazil’s election monitors flag hundreds of AI deepfakes despite labelling rules
Electoral rules require synthetic content to be labelled and ban deepfakes in campaigning; one observatory flagged 920 suspicious items in two months, 60% of them deepfakes, mostly unlabelled.
Misto Brasil ↗California orders independent verification of AI models and a committee on agent kill switches
In his final executive orders, Governor Newsom directed independent verification of frontier models, created an expert committee on shut-down mechanisms for AI agents, and asked for regulatory recommendations by 15 November, building on the state’s 2025 frontier-AI law.
Al Jazeera ↗Anthropic signs a 20-year lease for its first Canadian data centre in Nova Scotia
The company agreed to take capacity at EverWind Fuels’ Point Tupper campus, a project valued at about C$10 billion, marking its first dedicated compute footprint in Canada.
The Star (Malaysia) / Reuters ↗Hogan Lovells Cadwalader adopts two legal-AI platforms firmwide and creates an AI board
Rather than pick one vendor, the merged firm signed both Harvey and Legora and set up a board to govern build-versus-buy decisions, a pattern other large firms are watching.
Legal Technology ↗Jefferies expects consensus earnings growth for AI stocks to fall sharply by 2028
A broker note projects growth decelerating from near-triple-digit rates this year to the low twenties by 2028, arguing valuations assume a longer boom than the numbers support.
IANS Live ↗UN human-rights chief warns the window for AI regulation is closing
The UN High Commissioner for Human Rights urged governments to move from principles to binding rules on AI, warning that the pace of deployment is outrunning safeguards for privacy, expression and due process.
UN News ↗Russian envoy says Moscow and Washington are discussing joint AI data centres powered by small reactors
Kirill Dmitriev described talks on joint projects using small nuclear plants; US export controls on advanced chips to Russia remain in force, so any deal would face significant legal hurdles.
Startup Fortune ↗Microsoft walks away from Nscale’s Essex data-centre deal after grid delays
Microsoft withdrew from an agreement to anchor a UK site and rent roughly 23,000 Nvidia chips after power-grid capacity slipped into the 2030s, a reminder that electricity, not silicon, is the binding constraint.
FourWeekMBA (reporting The Telegraph) ↗India’s Zen Technologies buys a controlling stake in battlefield-AI developer Cogknit
The defence-training company’s board approved acquiring about 52% of Cogknit Semantics for roughly $14 million, extending its move into AI for military applications.
Business Upturn ↗South Korea’s DeepX starts mass production of its DX-M1 edge AI chip
The startup began volume production on Samsung’s 5 nm process and plans to tape out a 2 nm successor this month, targeting on-device inference rather than data centres.
KuCoin News ↗MapQuest adds a crowdsourced layer showing surveillance cameras and licence-plate readers
The map now optionally displays outdoor cameras from eleven vendors, including automated plate readers, giving the public a view of the AI-driven surveillance infrastructure around them.
The Next Web (via AI Weekly) ↗Report: Amazon tool scores UK warehouses by likelihood of unionisation
openDemocracy reports that an internal system called Atlas ranks sites on union risk using signals including survey answers and public posts, raising questions under UK data-protection and labour law.
openDemocracy ↗NIST, NSA and CISA publish practices for AI agents that act outside instructions
Ten voluntary practices cover limiting agents’ access to sensitive systems, continuous monitoring, rapid shutdown and automatic termination when controls fail.
Fox News live coverage ↗Massachusetts weighs AI rules that would be among the strictest in the United States
A state bill bundled with economic-development measures would impose disclosure, testing and accountability duties on AI developers and deployers that go further than most state laws to date.
WBUR ↗Nvidia’s share of China’s AI-chip market shrinks as Huawei closes the gap
A new report finds export controls and domestic substitution have cut Nvidia’s position in China, with Huawei’s accelerators taking a growing share of new deployments.
Yahoo Finance ↗ByteDance reported close to a MediaTek pact for AI data-centre chips
The TikTok owner is said to be nearing an agreement with MediaTek for custom accelerators, part of a broader move by Chinese platforms to secure supply outside US-controlled channels.
Yahoo Finance ↗Power shortages push AI data centres toward on-site generation
Supply-chain reporting shows operators ordering their own generation and storage equipment as grid connections lag demand, shifting capital from chips to electricity.
DIGITIMES ↗AI and chip stocks fall as a hot inflation print raises the odds of rate rises
Shares of AI chipmakers and data-centre operators dropped after inflation data strengthened bets on tighter policy, extending a sell-off that has hit memory and infrastructure names hardest.
CNN Business ↗Canada adds $500 million to its Regional Artificial Intelligence Initiative
The top-up brings the programme to $700 million over five years to help businesses adopt AI and Canadian AI firms scale; southern Ontario applications run to 19 January 2027.
Government of Canada ↗World items link to the original reporting and open in a new tab. ELP GPT does not reproduce articles; if a summary misstates a report, use the contact form and we will correct it.
Ask ELP GPT about any of these
Open the assistant and paste a headline; live search is on by itself for current topics.