Topic coverage
Dated GeoSignal briefings on Technology & AI. Read developments, assessment and source references in their original context.
The US Treasury Department has proposed COINS Act rules requiring notification or prohibiting specified US investments involving covered foreign persons in strategic technologies, but the framework remains open for comment and is not yet effective.
China has imposed immediate pre-export registration for specified AI, biomedicine and photovoltaic technology transfers while consulting on a broader revision of its technology export-control catalogue through November 15.
Participating EU member-state ministers adopted the Dublin Declaration on 14 October, aligning politically on trustworthy AI, European computing capacity and reduced strategic dependencies without creating new legal obligations or funding commitments.
The UK has expanded its frontier-AI cyber-defence programme across government and begun planning for serious incidents involving autonomous AI agents, while possible statutory reporting measures remain under consideration.
The UK is expanding frontier-AI use across government cyber defence after pilot programmes identified more than 1,000 potential vulnerabilities, while considering AI-incident safeguards within its cyber-resilience legislative regime.
Singapore is consulting companies and monitoring whether foreign controls on AI chips and remote computing could affect its data-centre and cloud-services sectors, while no disruption or mitigation programme has been confirmed.
Seven Dutch security and government bodies warned on 24 September that widely accessible AI is speeding cyberattack preparation and called on public- and private-sector leaders to strengthen baseline defences and oversight.
Alibaba has outlined a China-based AI-compute expansion centred on its Zhenwu V900 accelerator, but commercial release is planned for the first quarter of 2027 rather than reflecting a current deployment change.
Alibaba has set out a roadmap combining its Zhenwu V900 AI chip, a larger-model target and further data-centre expansion, while the chip’s reported commercial release remains planned for the first quarter of 2027.
Anthropic said on 22 September it had disrupted a suspected Russian state-linked cyber-espionage operation that used AI-assisted workflows against Ukrainian and European government, defence, diplomatic and drone-sector targets.
Anthropic said on September 19 that it banned accounts allegedly using Claude for China-linked military, intelligence, cyber and surveillance work and Russia-linked drone-software and dual-use procurement activity, while sharing information with authorities and industry partners.
Anthropic says it disrupted malicious use of Claude through August 2026, including AI-assisted cyber, surveillance, influence and weapons-related work linked at varying confidence levels to actors associated with Russia, Iran, China and other theatres.
AI governance is expected to feature when US President Donald Trump and Chinese leader Xi Jinping meet in Washington next week, but no bilateral safety agreement, shared standard or implementation mechanism has been announced.
Anthropic’s proposal for embedded independent evaluators, publicly backed by OpenAI, advances voluntary oversight of frontier AI systems but leaves common standards and cross-border enforcement unresolved.
China’s Foreign Ministry on September 14 rejected calls to sustain restrictions on China’s access to leading AI chips and chipmaking equipment, underscoring a widening public dispute over AI governance and strategic technology competition.
As of September 12, 2026, U.S. policy relies on targeted advanced-computing export controls and transaction-specific licence conditions after the January 2025 AI Diffusion Rule was rescinded before it took effect.
A September 8 joint US advisory alleges that several China-based AI companies conducted industrial-scale extraction from US frontier models; Beijing rejected the claims on September 9, and no new binding controls were announced.
The International Telecommunication Union is holding a 7 September workshop in Chongqing to identify security, trust, governance and evaluation priorities for future work on standards for agentic AI.
As of 23 August, US case-by-case licensing is allowing limited Nvidia H200 exports to approved Chinese customers, while Nvidia has denied it plans a China-specific language-processing chip.
The United States has expanded its prosecution of alleged members of Iran-based Mabna Institute, adding eight defendants to a case over a cyber theft campaign prosecutors say ended in 2017.
Taiwan confirmed on 13 August that a July overseas-origin cyberattack using AI agents targeted government agencies, with affected bodies having completed incident response.
The EU’s 21st Russia sanctions package has applied across member states since 24 July, imposing tighter dual-use and advanced-technology export restrictions on 51 entities linked to Russia’s military-industrial supply chains and sanctions circumvention.
Since 2 August 2026, the European Commission can enforce EU AI Act obligations on providers of general-purpose AI models sold in the EU, including through information requests, evaluations, corrective measures and fines.
As of August 5, 2026, U.S. export-control guidance confirms that licences remain required for covered advanced-computing items supplied to China- or Macau-headquartered entities, or entities ultimately owned there, even when they operate in third countries.
