Zürcher Nachrichten - Weapons, spyware and AI scams: Anthropic exposes Claude misuse

Here's a look at some of the cases, which have added to mounting worries over risks posed by rapidly advancing artificial intelligence.

- Spyware in Mali -

A tech consultant in Bamako, working with junta-led Mali's state intelligence service (ANSE), used Claude Code to build "Lakana 360" -- a surveillance system capable of monitoring roughly 25 million SIM cards across the country's mobile network.

The app, which Anthropic could not disable, can capture calls, texts and voice traffic, and even identify people by their voice if they switch SIM cards.

It can also spot who's using encryption or a VPN.

- Kamikaze drones in Russia -

A small team of likely freelance developers in Russia used Claude to build software intended to power autonomous "kamikaze" drones.

The drone swarm was programmed to pick its own targets and crash into them to detonate, without a human making the final call.

In their programming, the team repeatedly picked a location in Ukraine's Donetsk region as their target and used VPNs to circumvent Anthropic's geographic blocks.

The developers claimed to have received funding from Russian government and military research programs, though Anthropic assessed the group as a small freelance team rather than an official operation.

- Missile-making in Yemen -

A group in northern Yemen used Claude Code to help build weapons.

The popular coding tool helped design software for a guided rocket, drew up plans for a long-range missile, and helped develop an even faster type of missile called a hypersonic glide vehicle.

To avoid getting caught, they split the work across separate conversations with the AI.

Anthropic says there's no proof they ever finished a working weapon, but the group did test-fire one rocket. It apparently failed, because they went back to the AI afterward asking it to help figure out what went wrong.

- Propaganda in France -

Anthropic traced a fake news operation to LKM Company, a French advertising agency.

Using Claude, it mass-produced political articles across about 70 fake news websites in roughly 20 languages -- publishing at least 8,913 stories.

Each site was given a made-up journalist's byline and a fake Twitter/X account, boosted by hundreds of fake accounts posting comments meant to look like real people.

The operation flipped its message depending on who was paying, selling what Anthropic called "propaganda as a service."

It targeted the US, Brazil and France, and did a lot of work on the Democratic Republic of Congo, generally favoring the DRC's side in its conflict with Rwanda.

- Influencing in the UAE -

A UAE-linked influence operation targeted the Muslim Brotherhood and United Nations experts critical of Abu Dhabi's alleged role in war-ravaged Sudan.

The operation used Claude to generate reports, advocacy materials, social media content and testimony on Sudan.

Anthropic said the operation centered on an AI persona known as "Deadshot," guided by a command calling for "a coordinated transatlantic and regional operation to dismantle the Muslim Brotherhood globally."

The operation also ghost-wrote official testimonies intended for delivery at the 62nd session of the UN Human Rights Council, drafting them under "specific restrictions" that the speeches not mention the UAE.

- AI dates in China -

A China-based app studio built a network of more than 20 dating apps that claimed to be staffed by real people but were secretly run by Claude.

Over just two weeks in April 2026, Anthropic found more than 4,700 different Claude-powered AI characters chatting with at least 25,000 real users, sending about 2.36 million messages in total.

To seem more believable, real human workers were mixed in with the bots -- roughly one human for every three AI personas -- handling things the AI couldn't, like video calls.

E.Schneyder--NZN