Previous Columns

Beijing's destroyers, OpenAI's losses, poisoned npm packages, and Hong Kong's data cuts all ran in the same week, and the argument lives in their connection rather than in any single story.

The US suspended Fable 5 on safety grounds this spring and ordered Anthropic to cut foreign nationals off its systems, while China completed a decade-long infrastructure compromise campaign that the debate never addressed.

State-level compute control is the single factor expressing through four symptoms this week: infrastructure colonization, safety-pledge theater, scarcity-driven chokepoint leverage, and Hong Kong's simultaneous capital inflow and talent bleed.

Nvidia bets $150 billion on Taiwan's permanence, Anthropic nears a trillion-dollar valuation, and this week's defense cuts, agent security flaws, and autonomous intrusion document the fragility underlying both.

Taiwan's defense cuts land as Beijing's Scarborough patrols escalate; Anthropic reaches toward a trillion-dollar valuation as its agents prove demonstrably vulnerable. Both stories price commitment ahead of capacity.

Hong Kong's new licensing regime for stablecoin issuers went live this quarter. The same settlement system it regulates is the one organized criminal networks have quietly adopted for laundering and ransom payments, and the rules were not written for that.

Twenty-nine PLA aircraft, severed cables, and warships near Penghu mark a week when the Taiwan question moved from signal to sequence; Hong Kong's compliance sprint is reading the same calendar.

Autonomous systems now find vulnerabilities faster than institutions patch them; Hong Kong's financial sector, freshly regulated and newly exposed to digital asset infrastructure, is absorbing the first real test.

The SFC's investigation into broker system breaches is methodical, appropriate, and already beside the point; three structural problems are deepening while the regulators read last year's incident logs.

Three incidents in one news cycle confirm what the investment numbers already implied: AI infrastructure is now load-bearing, and the adversary adjusted long before most security teams did.

Taiwan Strait drills, South China Sea live rounds, a Hong Kong ransomware attack, and a cyber law rewrite in one week: when everything moves at once, single-domain risk thinking becomes a liability.

A ransomware seizure in Hong Kong, live fire near Manila, AI-accelerated code auditing, and a research deal at the HKMA: this week confirmed that frameworks without visibility are not frameworks.

Financial institutions racing AI from pilot to production are being briefed on competitive advantage while the same models they are deploying are simultaneously being turned against them.

Financial institutions are deploying AI as a productivity engine while adversaries run the identical capability against them; the industry's risk function has yet to register these as a single story.