Fencing off compute, a revolving door between the Army and AI, and a saturation attack on the grid

A ratio for new contracts, delivered without paperwork

According to the FT, which cited five people familiar with the matter, the condition applies to new contracts and leaves existing agreements untouched. Nvidia is said to have communicated it verbally through its sales force and regional teams rather than in writing, and this is described as the first time the company has imposed such a restriction on partners. The reports do not say whether the '51% non-Chinese' threshold is measured by revenue, capacity or contract count, or what happens on a breach.

Nvidia's statement did not address the report's substance. It said that 'datacenters and clouds built on American chips are the foundation that American model and application makers need to compete worldwide,' adding that 'longstanding U.S. policy encourages every cloud to build on American infrastructure and to win business everywhere they can.' The wording avoids a denial while folding the measure into US industrial policy.

A corporate line drawn where export controls stop

US chip controls on China have focused on exports of advanced chips and on transshipment through third countries, while the renting of compute to Chinese firms by overseas cloud providers has been flagged repeatedly by regulators without being codified. Earlier this year Super Micro employees were charged with smuggling about $2.5 billion of Nvidia AI servers into China, and an investigation into Megaspeed continues in Singapore. A cap on partners' customer mix reads as an attempt to narrow that channel through contract terms.

Beijing banned foreign AI chips from state-funded data centres a year ago, so Nvidia's position inside China had already shrunk. The new condition therefore bites less on mainland sales than on Chinese firms' use of third-country clouds in Southeast Asia or the Gulf. In Taiwan's legislature on 2 November, DPP lawmakers proposed amendments widening export controls on chipmaking equipment and materials to China, and the Ministry of Economic Affairs voiced support.

Beijing's reserved reply and what remains unknown

At the 3 November regular briefing, foreign ministry spokesperson Mao Ning told AFP she was 'not familiar with the specifics' and that 'all parties should engage in cooperation based on the principles of fairness, transparency and non-discrimination.' In the same briefing, speaking of the World Internet Conference, she said China opposes 'erecting digital iron curtains.' It is a formulaic reply that avoids escalation while framing the US measure as discriminatory. The questions ahead are whether Nvidia puts the condition in writing, whether the Commerce Department pursues know-your-customer rules for clouds, and how partner neoclouds disclose their contract mix.

120 jet drones launched alongside ballistic and cruise missiles

In the Air Force's breakdown, the 289 drones comprised Shahed-type craft (120 of them jet-powered), Gerberas and 'Parodiya' decoys; the 59 cruise missiles were Kh-101, Kh-59, Kh-69 and Iskander-K. The count of ballistic and Zircon missiles was left out of the Air Force statement; a Ukrainian outlet citing Espreso TV put the ballistic figure at 17. Neutralisations were 257 drones (89 jet-powered), 49 cruise missiles (three under clarification) and three ballistic or Zircon missiles; Interfax-Ukraine computed interception rates of 89% for drones and 83% for cruise missiles.

Launch points were given as Rostov, Voronezh and Kursk for ballistic missiles, Vologda, Orenburg and Oryol for cruise missiles, and Oryol, Kursk, Millerovo, Primorsko-Akhtarsk, occupied Donetsk and Hvardiiske and Chauda in Crimea for drones, with the main axes toward Kyiv, Kharkiv, Vinnytsia and Dnipropetrovsk regions. Which ballistic missiles penetrated the defences and where they struck was left unstated. All figures are Ukrainian claims with limited independent verification.

Targets shift to the grid and substations around nuclear plants

DTEK said three of its thermal power plants were hit and halted generation, and the Energy Ministry counted more than twenty attacks on energy facilities in ten regions within 24 hours. Kyiv's mayor Vitali Klitschko ordered emergency outages in the city and region 'due to damage caused by enemy attacks on facilities of the Kyiv regional energy hub,' with water supply and electric transport also disrupted. Ukrenergo CEO Vitalii Zaichenko told the Kyiv Independent that the attack included one of the largest waves of Russian ballistic missiles against critical infrastructure, and a substation critical to a nuclear plant was reported hit.

President Zelensky said Russia had changed tactics this year to target transmission and distribution networks as well as generation, and told the Berlin Security Conference that 'no European country can single-handedly defend its airspace against the number of ballistic missiles' and 'no country can build an effective drone defense system on its own.' Prime Minister Shmyhal stressed the threat to nuclear safety across Europe and asked partners for an additional €450 million on top of the roughly €800 million in the Energy Support Fund. On 2 November, an aid distribution point in Komyshuvakha, Zaporizhzhia region, was hit by successive aerial bombs; by 3 November, 31 deaths and about 50 missing were reported.

Daytime waves nine-tenths jet-powered; the defence load stays uneven

Launches continued after the night wave. The Air Force said that between 06:30 and 19:00 on 3 November, 81 drones arrived, 74 of them jet-powered, and 62 (56 jet-powered) were shot down or neutralised. Together with the 193 drones of the night of 1–2 November (75 jet-powered) and the 171 of the daytime on 2 November (about 60% jet-powered), the pattern is that Russia mixes large numbers of piston-engined craft and decoys at night and leads with jets by day, probing the defenders' capacity against fast targets. Interception rates for jet drones were 74% at night and 76% by day, below the overall figures that include propeller types.

The brief of the new post, in Driscoll's words

According to the release, he will 'work with OpenAI teams and leaders across governments, industry, academia, and allied institutions to advance practical modernization initiatives that support the buildout of democratic AI,' and help internal teams 'understand the operational needs of global national security institutions.' Driscoll said national security requirements 'are changing faster than at any time in the past half century,' and that AI would let the military 'keep pace by cutting through paperwork, making sense of more information, and getting the right answer to the right person faster.'

CEO Sam Altman said he was excited to welcome Driscoll 'to strengthen American leadership' and to work with other countries to 'spread the adoption of democratic AI.' A brief that foregrounds allies and responsible use signals a government business widening from deals with the US government toward allied markets. Driscoll himself said he was joining 'to help the United States and our allies put this technology to work responsibly, for the people who keep us safe.'

From Army transformation to an AI company

As Army secretary, Driscoll drove the Army Transformation Initiative, the Army's drone and counter-drone marketplaces, and the procurement overhaul known as Operation Jailbreak. He resigned in late August after months of reported strain with Secretary Hegseth, and has since advised the counter-drone company Perennial Autonomy and the public-affairs firm Watchtower Strategy. The move is a revolving door: the official who led the mass buy of uncrewed systems joins an AI company that sits among the department's suppliers.

OpenAI's government business after Anthropic's exclusion

OpenAI won a prototype Other Transaction agreement worth up to $200 million from the Pentagon's Chief Digital and AI Office in June 2025, and approved ChatGPT models became available on the GenAI.mil platform in August this year. Breaking Defense reports that after the dispute in which the military sought unrestricted access to Claude and Anthropic refused, the department shifted its classified-network AI integrations toward other providers, including OpenAI. The contrast between a company that kept its distance from the Pentagon over usage terms and one that hires a former secretary to expand with allies is hardening into the shape of the defence AI market.

How a finding of non-compliance on all three criteria fell apart

The Navy issued the MUSV solicitation on 26 March 2026 under Other Transaction prototype authority, funded from the July 2025 reconciliation bill. It called for a production-ready boat carrying two 40-foot containers and a 25-tonne payload for 2,500 nautical miles at 25 knots in sea state 4. Seven bidders were chosen for at-sea testing in May; Saildrone and Blue Water Autonomy were left out. The Navy had rated Saildrone non-compliant on all three evaluation criteria, dropped two findings after the company protested, and on the third the court found that the Navy had applied a standard to Saildrone that it applied to no other bidder.

Blue Water Autonomy's protest, heard in the consolidated case, was denied as the court found its elimination reasonable. The court had rejected the government's jurisdictional motion on 29 October and ruled against it again on the merits. The heart of the decision concerns procedural fairness, whether the same criteria were applied equally to every bidder, rather than the technical merits of the boats.

Three production awards and $600 million on hold

On 29 September the Navy had awarded production agreements for ten MUSVs each, at roughly $40 million a boat, to Galliano Marine Services, HII and Saronic Technologies, and $15 million each to Leidos, PacMar Technologies and Sea Machines, which passed testing; Birdon did not pass. Under the ruling the Navy may spend no more than $1.5 billion on MUSV construction, leaving about $600 million frozen. Saildrone gets another chance at an at-sea demonstration, with no deadline set for the reevaluation. Both excluded firms remain eligible for the second MUSV solicitation published on 23 September.

Founder and CEO Richard Jenkins said that 'the future of naval warfare will be fought on the water, not in a courtroom,' calling the ruling 'a win for fair competition, for taxpayers, and for national security.' A Navy spokeswoman did not immediately respond to the New York Times, and the service had declined to discuss the litigation in October.

A judicial check on fast procurement under OT authority

MUSV is a flagship of the Navy's push to mass-produce uncrewed vessels, and OT authority is used to bring in commercial newcomers while skipping Federal Acquisition Regulation procedures. The ruling shows that equal treatment of bidders remains open to judicial review even under OT agreements, imposing a procedural constraint on reforms that prize speed. Construction by the three awardees continues within the $1.5 billion ceiling, so the delay to the programme as a whole is expected to be limited, but the outcome of the reevaluation could reshape the second solicitation.

The AI clause and the spring 2027 meeting

The statement's AI content runs to two sentences: 'Recognising the speed of development in this domain, we will expand the Trinity House Agreement's Cyber Lighthouse Project to incorporate Artificial Intelligence,' and 'Through collaboration on a small number of shared military use cases, we will deepen bilateral cooperation in capability development, experimentation and adoption.' Writing in the Frankfurter Allgemeine Zeitung before the talks, Streeting said that 'through joint testing, development and ultimately the introduction of AI in our armed forces we will deepen our cooperation and keep pace with those who seek to harm us.'

According to Euronews, government and industry representatives from both countries are due to meet in spring 2027, and no specific AI systems, programmes or budgets have yet been announced. The statement also directs officials 'to maintain this momentum through 2027.' Since the agreement was signed in October 2024, German companies have committed £800 million (about €956 million) to the UK defence industry, including Helsing's Plymouth factory building AI-enabled underwater gliders.

Deep precision strike beyond 2,000 km, with uncrewed systems alongside

The centrepiece of the statement is a 'landmark project' to deliver Deep Precision Strike capabilities with ranges over 2,000 km within the decade; AI sits beside it as a new line. On land the two sides will 'explore cooperation on future capabilities and uncrewed systems,' and in the air the bilateral 'Flight Plan' includes 'future uncrewed and collaborative combat capabilities.' At sea, an Implementing Arrangement on Sting Ray torpedo procurement was signed the same day, and German P-8A aircraft flew from RAF Lossiemouth on exercises. Deeper space cooperation follows last month's leaders' agreement. The two countries co-chair the Ukraine Defence Contact Group and referred to the 'Coalition of the Willing' security guarantees.

Bilateral axes as the fast lane for AI adoption in Europe

Cyber Lighthouse was created under the 2024 agreement for information sharing and exercises; grafting AI onto it chooses a path of bilateral experimentation and adoption that does not wait for consensus in NATO or the EU. Limiting the scope to a 'small number' of use cases points to interoperable results in specific functions such as target recognition or logistics rather than a general AI strategy. Together with the UK–US defence AI and autonomy partnership of September, a layered set of UK-centred bilateral arrangements is forming the front line of military AI adoption in Europe.

The questions: common authorisations, procurement rules, barriers to small firms

The study falls to the board's Permanent Subcommittee on National Security Innovation. The memo's questions cover the feasibility of common security authorisations; whether the DFARS and the Financial Management Regulation support such purchases; barriers facing small businesses and nontraditional contractors; vendor authorisation, network access, facility and infrastructure requirements; and how vendors access customer data across classified environments. The authority cited is Section 907 of the FY2026 NDAA.

The framing reflects a judgment that what holds back AI on classified networks is the regime of authorisation and contracting rather than model performance. The explicit question of vendor access to customer data goes to the heart of the dispute over usage terms: the conditions under which commercial AI firms touch data in classified environments for training or operations.

Organisational options and a warning against 'hidebound' structures

On organisation, the memo puts four options on the table: a new defence agency or field activity, integrating functions into an existing organisation, consolidating development within existing staff organisations, or improving current structures. It warns that 'organizational structures that are excessively siloed, hierarchical, or hidebound by cumbersome processes' may 'prove unequal to this imperative,' extending January's realignment that centralised oversight of the department's innovation offices under Michael. Since May the department has announced agreements with several AI and cloud companies for classified networks; turning that stack of individual deals into an institution is the work still ahead.

From a June foothold to the late-October transfers

According to the report, the attackers first targeted a Wallet DevOps engineer in June through a tampered Homebrew installation downloaded while installing an AI coding tool. Binance detected the anomaly, replaced the device and rotated credentials, but the exposure revealed the engineer's role and colleagues. In September another employee was lured through 'a fraudulent recruitment process, posing as company executives' into downloading a malicious coding assignment that infected his work computer.

On 20 October the attackers used that compromised machine and an internal AI tool to take control of two other engineers' devices, finding a container deployment permission on one. About nine days later they used the stolen credentials to deploy a container whose malicious script reached the wallet-signing systems and authorised the transactions. Binance says it has frozen or intercepted about $2.85 billion, replenished 98% of affected assets and reopened withdrawals in phases from 04:00 UTC on 3 November.

An internal AI tool became the instrument of lateral movement

Binance summarised the methods as 'social engineering by fake recruitment interviews, malicious software downloads, compromised internal AI tooling, endpoint compromise and an advanced transaction-authorisation method.' It has not disclosed what the AI tool was or how it was abused, but its role as the path from one infected device to other engineers' machines shows how the cross-cutting access of workplace AI becomes an attack surface. Remediation includes rebuilding the wallet system in a new environment, redesigning internal AI tools and access controls, replacing devices, decommissioning related credentials and tokens, expanding AI-based behavioural anomaly detection and commissioning an independent security review.

Co-CEO Richard Teng said 'this wasn't one mistake or one breach — it was a months-long campaign,' and that the company was publishing the full report 'because the lessons go far beyond Binance.' Binance also said it detected and prevented 57 suspected North Korea-linked attacks in 2026.

The FBI's attribution and Pyongyang's denial

On 31 October the FBI attributed the attack to TraderTraitor, a North Korean state-sponsored unit, saying law enforcement had observed the perpetrators converting stolen assets into ether and expecting further laundering through related technology. On 2 November Kim Yo Jong, sister of Kim Jong Un, rejected the FBI's claims as a 'grave political provocation,' and China's foreign ministry said the same day it was not aware of the situation. Attribution rests on the FBI's finding and private tracing (TRM Labs), set against Pyongyang's denial.