Fake account posing as ‘Bangladeshi Expat’ used in covert foreign intelligence operation

A social media account named @BanglaExpatSharjah has been surfaced. Initially, it seemed to represent a Bangladeshi expatriate residing in Sharjah, United Arab Emirates. However, this individual was nonexistent. The entire persona was created through the use of artificial intelligence, or AI.

US-based AI company Anthropic disclosed the information in a report published on September 10. The company regularly investigates whether its AI systems are being used for malicious purposes.

The incident carries particular significance for Bangladesh because many Bangladeshis work in the UAE, and many of them use online groups to stay connected with family members and colleagues. That makes the creation of a fake account posing as a Bangladeshi expatriate a potentially sensitive matter.

Who was behind it?

According to Anthropic, the operation was carried out by, or on behalf of, a company called S2T Unlocking Cyberspace. It is a private intelligence firm reportedly linked to Israel and Singapore. The account used in the operation was shut down in June. The company did not provide any comment to Anthropic.

The broader operation was primarily aimed at monitoring people in Iran and the Gulf region.

How was AI used?

AI was being used for two main purposes.

First, it was used to analyse posts made by ordinary people. The operators would feed 25 posts at a time into an AI system and ask it to determine what kind of person had written them, where the person lived and what their political leanings were.

People were categorised into six groups: urban, religious, military, young, expatriate and rural. The operators also identified whether individuals were supportive of or opposed to the government.

Using this information, they generated reports in Arabic that were formatted to resemble official documents. The reports described what people were thinking and what might need to be said to influence or change those views.

Second, AI was used to generate posts for fake social media accounts. More than 250 such accounts had reportedly been created in advance.

What is known about the Bangladeshi account?

The fake accounts were organised into several groups. The Bangladeshi account was placed in a group labelled “UAE expatriates.” The same group also contained accounts posing as Pakistani workers and several labour forums in the UAE.

Anthropic said these accounts were still in the preparation stage and were detected at the beginning of testing. The company does not believe they were deployed for actual operations.

In other words, the report does not say that the fake Bangladeshi expatriate account joined any Bangladeshi group or communicated with any Bangladeshi worker.

Previous allegations

S2T Unlocking Cyberspace had also come under scrutiny in 2023. A confidential promotional document belonging to the company was reportedly found among leaked documents from the Colombian military.

The document claimed that the company could create fake accounts, use traps to obtain passwords and remotely install malicious software on phones or computers. The methods reportedly included infiltrating private WhatsApp or Telegram groups, collecting lists of members and then targeting them individually.

Anthropic said AI-generated posts could make fake accounts appear more credible. However, the claims contained in the promotional document did not entirely match what Anthropic itself observed. Before the accounts were shut down, the company found no evidence that any real person had been harmed.

What is the situation now?

Anthropic said that surveillance, profiling people's political identities and operating networks of fake accounts are all prohibited under its rules. The company has also shared technical information with organisations responsible for monitoring such activities.

However, only one account was shut down. The report does not say whether the broader surveillance operation itself has been halted.