A cargo ship has been sunk by a drone off Bulgaria. Eleven sailors are missing. Meanwhile, the IMF says AI investment is helping hold up global growth—but a sharp reversal could make the next downturn worse. And in Nepal, the telecom regulator has fined 14 internet providers for breaching approved tariffs and packages.

The thread running through all three stories: when systems become powerful, accountability matters.

Two cargo ships were hit by drones off Bulgaria on Tuesday. One sank; 11 sailors were reported missing. The other, the Maltese-flagged Chelsea-3, was hit by three drones, but all 15 crew members were rescued.

The sunken vessel, HS Dignity, was sailing under a Guinea-Bissau flag. Five crew members were rescued, while the search for the other 11 continued. Bulgaria has not publicly established who was responsible. Prime Minister Rumen Radev said the attack's outcome benefited Russia and summoned the Russian ambassador; Moscow said it had no information about the incident.

Why it matters: this is not just a maritime-security story. The Black Sea is a key route for grain exports from both Russia and Ukraine. Repeated attacks can raise shipping risks, insurance costs and pressure on a trade route that matters to global food supply. NATO has said it will adjust its eastern-flank presence as needed.

The crucial detail is what remains unproven: suspicion is not confirmation. The investigation and rescue operation are still central to the story.

The Nepal Telecommunications Authority has fined 14 internet service providers Rs 50,000 each—a combined Rs 7 lakh—for violating approved tariffs, packages and service conditions. The regulator says some providers charged more than authorised rates, sold unapproved packages or failed to deliver services matching what customers had bought.

The companies have been directed to withdraw unauthorised plans and compensate customers who suffered losses. Among the providers named in the report are WorldLink, Vianet, Subisu and Classic Tech.

The consumer test: the fines matter, but they are not the finish line. The stronger test is whether providers correct their packages, deliver the speeds they advertise and return money to customers who were wrongly charged. Keep your package details, bills and speed-test records if you need to challenge a charge.

*The data presented below are from the official sites of NEPSE, NASDAQ and London Bullion Market Association respectively.

🇳🇵 NEPSE closed Tuesday at 2,599.93, up 4.12 points from Monday's 2,595.81.

🥇 Gold Rs 292,200 per tola on Tuesday, down Rs 2,200 for the day and Rs 4,300 over two sessions.

🥈 Silver Rs 4,355 per tola, down Rs 60.

The dispute is testing whether young protesters can turn street anger into sustained political pressure. For a democracy, the core question is straightforward: how can voter rolls be cleaned without wrongly excluding eligible voters?

Reuters reports that Anthropic and OpenAI have sought a say over what external evaluators can publish, citing concerns that reports could expose trade secrets. The risk is that commercially sensitive information can overlap with details needed to judge a model's safety. If labs can restrict access or redact too much, the public may not learn what the tests uncovered.

There is an important distinction between the companies' positions. OpenAI says it is not seeking approval rights over the full text of evaluators' reports and that its usual practice is to redact confidential information. Anthropic has said external evaluators should be able to publish key findings without editorial control, while some details of its evaluation arrangements remain under discussion.

Why it matters: a safety review is useful only if its findings can be trusted. Independent evaluators need enough access to do serious work—and enough freedom to report material risks.

Google has begun rolling out its Gemini 4 Argon model to paid Gemini app subscribers and developers. Google says it is built for tasks including coding, cybersecurity, finance and law. The company also says internal use helped engineers increase AI-generated code production by 35%; that is Google's own reported measurement, not an independent benchmark.

In one downside scenario, a sharp drop in AI investment and asset prices could reduce global GDP by 1.1% by 2028, including a 1.8% hit to US output. A prolonged Middle East war that kept oil above US$130 a barrel through the first quarter of 2027 could cut next year's growth by another half a percentage point.

These are risk scenarios, not the IMF's central forecast. Its baseline projects 3.4% global growth in 2027, assuming the Strait of Hormuz reopens by March and shipping conditions gradually return to normal.

The takeaway: AI is no longer just a technology-sector bet. Its infrastructure spending supports chipmakers, construction, energy demand and wider trade. That also means disappointment in expected AI profits could travel much further than a fall in a handful of tech stocks.

AI companies say outside safety reviews are important. Who should have the final say over what those reviews publish?

Reply with A, B or C:

Answers just below. No peeking. 👇

📖 Read: Reuters’ investigation into independent AI safety reviews — understand why transparency becomes complicated when companies control access to their own AI models.

🎧 Explore: The IMF’s latest global economic outlook — see how an AI investment reversal could spill into global growth, debt and financial markets.

📱 Follow: Nepal Telecommunications Authority — track regulatory notices, approved internet packages and consumer-protection updates following the fines against 14 internet providers.

Systemic risk (sis-TEM-ik risk)

Meaning: The risk that trouble in one company, market or institution spreads and threatens a much wider system.

Example: “The IMF is warning that an AI investment correction could become a systemic risk if falling valuations spill into debt, trade and the wider economy.”

For business/promotions, hit us up at: [email protected]

Made it this far? Reply and tell us what you want unpacked tomorrow — we read every message.