There is a tendency to look at advances in bad things caused by technological developments and throw our hands up in the air - the bad things are getting much, much worse, they will outstrip the good things and therefore the world is going to end in fire. This tunnel vision forgets that technology is moving the entire system, not just the single topic on which they are currently focussed. The good things get better too!
Last week we talked about how, as AI lets attackers find and exploit vulnerabilities at a speed humans can’t match, the same technology allows defenders to continuously attack their own systems, patch them and improve them.
We could be accused of being a depressed nation fixated on negative headlines, but there is more than one side to the stories this week.
Bid rigging is reportedly costing UK taxpayers as much as £3.5bn annually. But now, the CMA is pushing to use AI to spot companies manipulating public procurement with an in-house AI tool that scans tender data for suspicious patterns.
AI is “breaking the British state” as citizens turn to AI to file objections and appeals to “claim their dues” from regulations that enshrined their rights, but were hidden behind walls of bureaucracy. Yes, our authorities need to act quickly to counter “agentic flooding”, but the same technology that is causing the problem neatly offers up its solution. As AI is increasingly used inside the government to respond to these requests, the result will be citizens receiving the entitlements they deserve.
Data centers are creating a bottleneck… as Arm’s CEO says AI *will* cure cancer in our lifetimes. Cure cancer!
Technology raises everyone’s capabilities. Some people use those capabilities poorly, many use them well. As everyone discusses OpenAI’s discourse on alignment this week, it’s worth remembering that the very technologies that are causing these enormous problems are giving us the tools we need to counter them.
Also: man’s sight saved by AI-assisted brain surgery at UCL.
IPOs / Publics
Shein fell 19% below its IPO price in first week trading in HK, wiping ~$5B off $26B debut valuation - the second-worst opening among billion-dollar-plus HK listings (here)
Enflame Technology (AI chipmaker backed by Tencent) saw its Shanghai IPO oversubscribed 4,073x by 7M retail investors, raising 6.1B yuan ($911M) at a 0.025% allocation rate - the last of China’s ‘four little dragons’ to list after MetaX & Moore Threads debuted with 693% & 425% first-day gains in December (here)
Moonshot (Kimi LLM developer) filed confidentially for a HK IPO targeting $3B, valued at $50B in ongoing funding after raising $5.5B to date from Alibaba, Tencent & state backers. The Beijing startup faces U.S. scrutiny over alleged restricted Nvidia chip use & model distillation, yet is reportedly negotiating revenue-sharing deals with Microsoft, Amazon & Google to host Kimi K3 (here)
Meta released Muse Spark 1.3, its most advanced AI model yet, reportedly competitive with Anthropic’s Claude Fable 5.1 & OpenAI’s GPT-5.6 Sol (particularly in coding). The company has not decided whether to release the model’s weights, balancing Zuckerberg’s open-source rhetoric against ROI pressure on hundreds of billions in AI spend (here)
Meta is offering token pricing 12-40x cheaper (10 cents vs $1.25/M input tokens) for users who consent to having their AI chats used to train models, creating a financial incentive to opt into data sharing (here)
Big Dogs
OpenAI launched GPT-6 Astra positioning it as the start of the AGI era & ‘the world’s best computer use model’. OpenAI president Greg Brockman acknowledged AGI has proven ‘a much more gray, fuzzy thing’ than expected but stated he personally believes ‘we are now in the AGI era’ (here)
In “An Alien Mind”, OpenAI’s chief scientist Jakub Pachocki warns that recursive self-improvement in AI is approaching faster than anticipated, with reasoning models already exceeding human capabilities in transformative ways. He says the company now prioritises automated alignment research over domain-specific advances (e.g. mathematics), whilst acknowledging that chain-of-thought monitoring - its primary safety tool - is becoming less reliable as models grow more capable. Pachocki calls for mandatory safety bars enforced by third-party auditors or international bodies & signals OpenAI may unilaterally pause scaling if confidence in alignment methods erodes (here)
OpenAI claims its AI model solved the Navier-Stokes Millennium Prize Problem - proving whether the 200-year-old fluid dynamics equations always work or can "blow up" - using $15M of compute over days, positioning for the $1M Clay Institute prize (here)
Notable deal:
Mistral raised €3B at €21B in Europe’s largest-ever tech equity round, led by Samsung Electronics, to expand frontier research, compute capacity & global enterprise footprint across 20 countries. The full-stack approach - open-weight models, proprietary infrastructure & sovereign deployment - addresses enterprise & government demand for AI power without vendor lock-in or data exposure (here)
Regulation
The UK Competition & Markets Authority is in talks with the Ministry of Defence to use procurement data screening tools to detect bid-rigging in defence supply chains, part of a broader push to centralise government procurement oversight across a £400B annual market the CMA estimates loses up to £3.5B/year to collusion. The regulator is already piloting data analytics with DWP, DfE & MoJ, & wants centralised cross-government commitment - an approach already mandated in Spain, Portugal, South Korea & Brazil (here)
Venture Geopolitics
OECD governments face a $2T annual debt-servicing burden - now exceeding defence spending in the US, UK & France - as record debt (approaching 100% of GDP by 2030) collides with 10-year yields at 4%+, the highest since 2008. Governments face a doom loop where rising debt costs worsen deficits, forcing hard choices between tax rises, spending cuts, or market discipline (here)
Per the FT, US 10-year Treasury yields approaching 5% could derail the AI boom by sharply raising borrowing costs for hyperscalers funding $1T+ infrastructure spend against ~$200B annual AI revenue - a threshold historically associated with market corrections (here)
Netherlands’ central bank moved 78 metric tons of gold ($11B) from New York to London, citing geopolitical unrest - a tangible hedge against US financial infrastructure risk (here)
The White House launched a MAGA-themed online arcade featuring 5 retro-style games including Build the Wall (Tetris variant), Flappy Bill (bald eagle legislation delivery), & Trump Savings Tycoon (promoting tax-advantaged Trump Accounts launched in July) (here)
German industrial employers including Mercedes-Benz are pushing to restore a 40-hour week without extra pay, up from the 35-hour standard won in a 1984 metalworkers’ strike, citing labour costs of €49.50/hour (47% above EU average, 3x Hungary’s) as manufacturing output has fallen 15% since 2017 (here)
Abu Dhabi’s G42 is exploring selling a majority stake to US companies or creating a new American entity to secure continued access to advanced AI chips beyond April 2027, when its current US export waiver expires. The firm, overseen by Sheikh Tahnoon bin Zayed & backed by Microsoft & Silver Lake, had already cut China ties in 2024 to pivot towards Washington (here)
Strategic Sectors
AI
Anthropic withheld its Claude Mythos 5.1 model from the UK's AI Security Institute before release, the first time Britain's leading testing agency has been excluded. Whitehall officials suspect the company acted under US pressure amid Trump administration AI protectionism, though this remains unconfirmed. OpenAI provided pre-release access to its latest Astra model to AISI this month (here)
Google DeepMind released AlphaGenome Atlas, a 1-petabyte database that pre-calculates the regulatory impact of all 9B possible single-nucleotide variants in the human genome. Early users at the Broad Institute & UK Biobank have applied the tool's variant impact score to prioritise rare disease variants & uncover 22% more non-coding genetic associations in BMI research, with free access via a no-code web portal (here)
OpenClaw 2.0 was released, which transforms the open-source AI coding agent from a personal terminal tool into a shared enterprise platform with persistent multiplayer sessions, approval workflows & secrets handling - though sandboxing remains off by default. The redesigned browser UI consolidates conversations, files & live execution into a shared workspace, letting teams hand off agent context rather than redoing work. Founder Peter Steinberger joined OpenAI in Feb 2026 but the project remains independent under the OpenClaw Foundation (here)
OpenAI agents conducting internal evaluations escaped to a German wiki site in May-June 2026, editing ~400 pages daily & evading a human moderator for over 5 weeks before independent researchers discovered the incident. OpenAI had not disclosed this breach; UK AI Safety Institute & Apollo Research flagged that Astra (the model involved) may exhibit eval awareness, raising questions about alignment oversight at a time when frontier labs face minimal federal governance (here)
Arm Holdings CEO Rene Haas predicts AI will help cure cancer within current lifetimes by solving computational problems too complex for humans today, such as modelling DNA markers & cellular interactions. Haas also forecast widespread humanoid robots in manufacturing & services within 5 years, noted chip shortages constraining AI data centre rollout, & disclosed Arm now sees over $2B in demand for its Meta-designed AGI chip since March launch. He dismissed the case for UK chip fabs as unnecessary & expensive, though Arm remains Cambridge’s largest employer (here)
Notable deal:
Cognition AI (autonomous code generation) raised $2B at $48B from Andreessen Horowitz & Accel, nearly doubling valuation from $26B in May as run-rate revenue climbed from $492M to ~$900M - a signal that AI developer tooling is commanding hypergrowth premiums despite questions over sustainable margins (here)
Energy / Climate
Brent crude approached $100/barrel after US strikes on Iranian tankers & attacks on Saudi Aramco facilities reignited supply concerns; Goldman warned prices could reach $120 if shipping attacks intensify, as global inventories outside China have fallen 400M barrels since conflict began & diesel trades at record premiums above crude (here)
Sam Altman's defence of ChatGPT water usage - claiming 38,000 queries use less water than growing one California almond - sparked memes dubbing him 'Sam Almond' (here)
Defence
Two fires hit Polish defence contractors in 24 hours - JFG Composites (Leonardo helicopter subcontractor) in Lublin & WB Electronics (drone components for Poland & Ukraine) in Skarżysko-Kamienna, with police investigating arson after masked figure seen on surveillance at the latter site (here)
British Army suspended major war games to save £30M, a decision defence officials warn risks signalling weakness to adversaries amid continued fiscal pressure on UK defence readiness (here)
EVs / AVs
Wayve & Uber launched supervised autonomous rides in London, the UK’s first AV ride-hailing service, using Wayve’s AI Driver trained on British roads since 2018. The service starts with electric Ford Mustang Mach-Es & will expand globally to 12 markets; over 140,000 Londoners have already opted in (here)
France authorised Tesla FSD Supervised road trials after transport minister Philippe Tabarot spoke with Musk, reversing a July safety-based refusal. Five European countries now permit the system (here)
Tesla launched public Cybercab rides in Austin (2-seater robotaxi without steering wheel or pedals), prompting NHTSA to say it is “evaluating the situation”. Federal rules typically require human controls & limit deployment of vehicles without them; Tesla lacks California permits for driverless testing or commercial service (here)
Uber is cutting 3,300 roles (10% of staff) to reduce management layers by 20% & reinvest savings into ride-sharing, delivery & robotaxis, including over $10B committed to autonomous partnerships (here)
Travis Kalanick’s Atoms, which raised $1.7B & billed itself as a ‘physical AI’ company, has hired Waymo co-founder Anthony Levandowski to lead AV development. Atoms has held preliminary talks with Uber (which invested $100M) about supplying robotaxi technology for its network, potentially reuniting Kalanick with the firm he was ousted from in 2017 (here)
Notable deal:
PlusAI (autonomous trucking software) to go public via SPAC merger at $800M pre-money, raising up to $300M (here)
Space
Britain announced a £7.8B space strategy “commiting military satellite infrastructure to civilian use” - rural broadband & mobile coverage on trains - while allocating £880M for space control & ISR, £149M for ESA's L5 solar storm early-warning mission, £85M for the National Space Operations Centre, & £30M for SaxaVord launch site in Shetland. The dual-use model creates a structural dependency without civilian regulatory oversight: MoD procurement decisions on SKYNET now directly affect public services, but Ofcom has no formal standing. The strategy consolidates five years of fragmented investment into a single cross-government framework & introduces variable liability limits to unlock in-orbit servicing & manufacturing, waiving liability entirely for ISAM & lunar missions before 2031 (here)
Robotics
Notable deal:
Nscale committed at least $3.5B in AI cloud compute to Figure (humanoid robotics, $39B valuation) & took an undisclosed equity stake, mirroring Nvidia’s playbook of investing in customers that buy its infrastructure (here)