An analyst reviews a network of linked social media accounts on a monitor.

Malaysia Asks Anthropic About an AI Influence Campaign: What Happened

Key Takeaways

  • Malaysia is seeking further details from Anthropic about the alleged use of its Claude AI models to power a political influence operation targeting voters in the country, the Communications Ministry said in a written parliamentary reply on Tuesday 17 November 2026.
  • The request follows Anthropic's own threat report, published on 10 September 2026, in which the company said it had removed an account that used Claude to run a commercial election manipulation platform primarily targeting Malaysian voters.
  • According to Anthropic, the platform included about a thousand fake accounts on X, a fake news outlet, fabricated dossiers and a YouTube channel. It was built to profile voters across all 222 of Malaysia's federal parliamentary constituencies.
  • Anthropic attributed the operation to BBS Bilisim Teknolojileri, a technology firm based in Istanbul. The company denies the allegations.
  • The Malaysian Communications and Multimedia Commission (MCMC) will determine whether legal action needs to be taken once it has the details, and Malaysia is drafting a Bill to regulate AI usage.
  • Much remains unknown: whether Anthropic has responded, how many voters the campaign actually reached, who paid for it, and whether it affected any specific election. Anthropic itself found no evidence the campaign gained meaningful traction with real users.

What Malaysia Asked For

The story broke on Wednesday 18 November 2026, when The Straits Times, carrying a Bloomberg report, reported that Malaysia is seeking further details from Anthropic over the alleged use of its artificial intelligence models to power a political influence operation targeting voters in the country. The Malaysian Reserve carried the same Bloomberg report the same day.

The request came through Parliament. In a written parliamentary reply on Tuesday 17 November, the Communications Ministry told Parliament that the Malaysian Communications and Multimedia Commission (MCMC) has sought more information from Anthropic about its report. MCMC will then determine if legal action needs to be taken, the ministry said. The ministry also said Malaysia is drafting a Bill to regulate AI usage to ensure accountability and risk management.

The reply was given to Datuk Seri Dr Ronald Kiandee, the member of Parliament for Beluran, who had asked whether MCMC's investigation had identified the parties behind the manipulation operation and what actions had been taken against the fake account networks and fabricated news sites. The reply did not name the parties, and it did not say what action, if any, has been taken so far. It said, in effect, that the regulator is still gathering information.

That is an important distinction. Malaysia has not accused Anthropic of wrongdoing. It has not announced charges, fines or any enforcement action. It has asked the company that discovered the operation and wrote the report about it for more detail — the evidence behind the report, the technical specifics of how the operation worked, and whatever else Anthropic knows that did not appear in the published version. Only after receiving that information will MCMC decide whether legal action is warranted.

This is also not the first time the matter has come before Parliament. On 4 November 2026, the ministry gave a similar written reply to Ahmad Fadhli Shaari, the member for Pasir Mas, who had asked what steps the government would take to investigate foreign election meddling conducted through generative AI platforms, including Claude. In that reply, the ministry said MCMC was treating the allegations seriously and had sought detailed technical clarification and supporting evidence from Anthropic, and that enforcement action would be pursued under the Communications and Multimedia Act 1998 alongside the Online Safety Act 2025. The 17 November reply confirms that the request for details is still the active step, and that no enforcement decision has yet been made.

The two replies, two weeks apart, tell their own story. The government is moving deliberately, through its regulator and through existing law, rather than announcing dramatic action. Whether that caution reflects the difficulty of the case — a foreign company, a foreign AI provider, and an operation that may have had little real-world effect — or simply the pace of bureaucracy is something only the ministry can answer, and it has not done so yet.

What Anthropic Said Happened

To understand what Malaysia is asking about, it is necessary to go back to 10 September 2026, when Anthropic published a threat intelligence report covering misuse of its Claude models that it said it had disrupted between December 2025 and August 2026. The 154-page report covered cyber operations, influence operations, surveillance, scams and fraud. The most prominent case in it was a commercial election manipulation platform that primarily targeted voters in Malaysia.

According to Anthropic, the operators used Claude to build a system capable of profiling and targeting Malaysian voters by constituency — all 222 of the country's federal parliamentary constituencies. The system was designed to target what Anthropic described as some of Malaysia's most sensitive political and social fault lines: race, religion and royalty. The campaign, according to reporting on the Anthropic document, focused on sensitive social issues including royalty, religion and race.

The machinery of the operation, as Anthropic described it, had several parts. The platform managed about a thousand fake accounts on X, the social media platform formerly known as Twitter. It ran a fabricated news site called "Malaysia Pulse" — Anthropic identified the domain malaysiapulse.com and said it was registered on 10 May 2026 — and a YouTube channel. It produced fabricated intelligence dossiers, documents designed to look like genuine intelligence material. And it harvested demographic census records and voter data to build profiles of electorates constituency by constituency.

Anthropic labelled the case GTG-84005 in its report, and categorised it as a "Category Two" threat, meaning it was active across multiple platforms but failed to gain significant traction within genuine online communities. That last point matters. According to Anthropic, the campaign showed no evidence of gaining meaningful organic traction among genuine users. The fake accounts existed, the fake news site existed, and the dossiers were written — but there is no evidence in the published material that real Malaysian voters were persuaded by them in significant numbers.

The report also contained a detail that raises as many questions as it answers. The operators, Anthropic said, pursued a contract with Malaysia's national communications regulator — the MCMC itself — and requested one million artificial views for the Prime Minister's social media account. Anthropic found no evidence that the government was involved or that the contract was ever granted. The detail is strange, and it is unexplained: why would an influence operation aimed at manipulating voters also seek a contract with the regulator it was supposedly targeting? Anthropic's report does not answer that question, and neither does any public reporting since.

Anthropic attributed the operation to BBS Bilisim Teknolojileri, a technology firm based in Istanbul. The company denies the allegations — it denied involvement on 11 September 2026, the day after the report was published. Anthropic said it has banned the offending Claude accounts. The company did not identify who paid to use the platform for the Malaysia-focused activity, or whether the operation was connected to any particular Malaysian election.

That attribution is the most contested part of the story. Anthropic's word is the only public evidence linking the operation to BBS Bilisim Teknolojileri. The company denies it. No government or law enforcement agency has publicly confirmed the attribution, and no independent investigation has been published. Readers should treat the attribution as Anthropic's claim, not as an established fact — and the same applies to every detail of the operation that comes only from Anthropic's report.

Why Malaysia Is Asking Now

The timing of Malaysia's request is worth examining. Anthropic's report was published on 10 September 2026, more than two months before the parliamentary reply of 17 November. What happened in between?

According to The Edge Malaysia, MCMC said after the release of the Anthropic report that it was looking into the allegations. But the same report notes something striking: MCMC had also said that such manipulation activities were not part of its mandate. That apparent contradiction — a regulator looking into allegations that it says fall outside its mandate — has not been explained publicly, and it may help explain why the ministry is now emphasising coordination with enforcement agencies and AI platforms, and why it is leaning on the Communications and Multimedia Act 1998 and the Online Safety Act 2025 rather than on any single regulator's powers.

The parliamentary route also matters. The 17 November reply came in response to a question from an opposition member of Parliament — Kiandee sits with Perikatan Nasional — and the 4 November reply came in response to another opposition member. The government is answering questions put to it in Parliament, not volunteering information. That is normal parliamentary practice, but it means the public record of what Malaysia knows and is doing consists largely of short written replies to specific questions, not detailed public statements. The brevity of those replies is part of why so much about this case remains unclear.

There is also a legislative angle. Both parliamentary replies mention that the Digital Ministry is drafting an Artificial Intelligence Governance Bill to strengthen AI accountability, and the 18 November reporting describes it as a Bill to regulate AI usage to ensure accountability and risk management. The Bill is not yet law, and its contents have not been published. Its existence, however, shows that the Malaysian government sees the Anthropic case as evidence of a gap in its current law — a gap it intends to close with new legislation aimed at AI developers and deployers. Whether the Bill would have changed anything about this case, or whether it would give MCMC powers it currently lacks, is not yet possible to say.

What the Case Means for AI and Elections

The Malaysia case is the clearest public example yet of a general-purpose AI model being used to run a multi-part influence operation. Earlier cases of AI misuse in elections tended to involve a single trick: a fake voice call, a doctored image, a batch of generated text. This case, as Anthropic describes it, involved a whole pipeline — voter profiling from census and electoral data, fake social media accounts, a fake news outlet, fabricated dossiers, and a YouTube channel — all built with the help of one AI model.

That is a significant development, and it is one that should worry anyone who cares about elections. The cost of running an influence operation has fallen dramatically. A thousand fake accounts, a convincing fake news site and a set of professional-looking dossiers once required a team of people and a substantial budget. Anthropic's account suggests that a small group with access to a frontier AI model can now produce the same material far more cheaply and quickly. The barrier to entry for election manipulation has fallen, even if the effectiveness of the manipulation has not yet been proven.

But the case also shows the limits of that threat, at least for now. Anthropic's own assessment is that the operation failed to gain meaningful traction with real users. Fake accounts and fake news sites are only useful if real people read them, share them and believe them. There is no public evidence that happened here at scale. The operation was detected by the AI company whose model it abused, the accounts were banned, and the report was published. The system of detection and disclosure worked, at least in this instance.

The case also raises a question that goes beyond Malaysia: who is responsible when a general-purpose AI model is misused in this way? Anthropic detected the misuse, banned the accounts and published a report. That is more than many AI companies have done in similar situations. But the report was published two months before Malaysia's regulator formally sought more detail, and the detail that would allow an investigation — the technical evidence, the account data, the specifics of who did what — is held by a private American company, not by the Malaysian authorities. The regulator's request is, in effect, a request for the evidence behind the report. How quickly and how fully Anthropic provides it will shape whether any enforcement action is possible.

That question is not unique to Malaysia. Anthropic's September report also detailed alleged influence operations in other countries, including Russia, China and Iran. The pattern is the same in each case: a private company discovers misuse of its own product, publishes what it found, and leaves governments to decide what to do next. Governments, in turn, are discovering that their existing laws were written before AI models existed, and that they may not have the tools to investigate or prosecute this kind of operation — especially when the operators, the AI provider and the targets are in different countries.

Malaysia's answer, for now, is to ask for more information, to coordinate between agencies, and to draft new legislation. It is a measured response to a case that is, by Anthropic's own account, already over: the accounts are banned, the operation is disrupted, and there is no evidence it changed any votes. The question is whether the next operation will be caught as quickly, and whether the law will be ready when it is.

Readers interested in how AI misuse is detected and policed more broadly can read our earlier coverage of the Japan-led probe into AI-driven cyberattacks and our guide to AI safety, its key risks and how to use AI safely. For a different kind of Claude misuse — one with a very different outcome — see our article on how Anthropic's Claude submitted a fake police tip.

What Happens Next

The immediate next step is straightforward: Anthropic must decide how much detail to give MCMC, and MCMC must decide whether what it receives justifies legal action. Neither side has said anything publicly about that exchange since the parliamentary reply. The Bloomberg report that broke the story on 18 November is brief, and it does not say which Malaysian authority made the request, what specific information the government wants, or whether Anthropic has responded.

There are several possible outcomes. MCMC could conclude that the evidence supports action under the Communications and Multimedia Act 1998 or the Online Safety Act 2025 — but action against whom, and in which jurisdiction, is far from clear. The alleged operator is a Turkish company that denies the allegations. The AI provider is American. The fake accounts were on X, an American platform. The fake news site used a domain registered through the global domain system. A Malaysian prosecution would face all the usual difficulties of cross-border enforcement, compounded by the fact that the underlying conduct — using an AI model to write content and run fake accounts — may not clearly be an offence under laws written before such tools existed.

Alternatively, MCMC could conclude that there is nothing further to do — that the operation was disrupted, the accounts banned, and no Malaysian law was broken in a way that can be prosecuted. That would be an unsatisfying outcome for anyone who sees the case as an attack on Malaysia's democratic process, but it may be the realistic one. The ministry's own earlier statement, that such manipulation activities were not part of MCMC's mandate, suggests that the regulator itself is uncertain about its powers here.

The third possibility is that the case becomes the catalyst for the AI Governance Bill, and that the Bill gives Malaysian regulators clearer powers for the next case rather than this one. That is often how regulation works: a specific incident exposes a gap, and the law is changed to cover the next incident, not the last one. If that is what happens, the Malaysia case will be remembered less for what was done about it and more for what it revealed about the state of AI regulation in 2026.

For readers in other countries, the case is a preview. The same model that was used to target Malaysian voters is available to anyone with an internet connection and a payment method. The same techniques — fake accounts, fake news sites, fabricated dossiers, voter profiling — can be aimed at any electorate. The fact that this operation appears to have failed is reassuring, but it is not a guarantee. The next operation may be better run, better targeted, and harder to detect. The question for every country holding an election in the next few years is whether its regulators, its laws and its AI providers are ready.

One practical lesson is already clear: detection depends on the AI companies themselves. Anthropic found this operation because it monitors how its models are used and investigates suspicious activity. Not every AI provider does this to the same standard, and not every provider publishes what it finds. The transparency of Anthropic's September report — whatever its limitations — is what made Malaysia's request possible. Without it, the operation might never have been known about at all. That argues for stronger expectations, and possibly stronger legal requirements, on AI providers to monitor for misuse and to disclose what they find.

What We Could Not Verify

  • Whether Anthropic has responded to MCMC's request. No public statement from Anthropic or from the Malaysian government confirms whether the company has provided the additional details, or when. The 17 November parliamentary reply says only that MCMC has sought more information.
  • How many Malaysian voters the campaign reached. Anthropic said the operation showed no evidence of gaining meaningful organic traction, but no figure has been published for how many real users saw, shared or believed the campaign's content. The scale of its actual impact is unknown.
  • Who paid for the operation and whether it was connected to any particular election. Anthropic did not identify who funded the platform or whether it was tied to a specific Malaysian election. The description of the operation as "commercial" raises the question of who the customer was, and it has not been answered.
  • The attribution to BBS Bilisim Teknolojileri. The attribution comes solely from Anthropic's report. The company denies the allegations, and no government or independent investigation has publicly confirmed them.
  • Why the operators sought a contract with the MCMC and requested one million artificial views for the Prime Minister's social media account. Anthropic reported both details but did not explain them, and no other source has clarified them.
  • The contents and timetable of Malaysia's draft AI Governance Bill. The Bill has been mentioned in parliamentary replies but its text has not been published, so its scope and powers cannot be assessed.
  • The Yahoo News Singapore version of the 18 November story. The development was verified through The Straits Times (carrying the Bloomberg report), The Edge Malaysia and Channel News Asia. A Yahoo News Singapore edition of the same wire story was not independently opened during this research, so it is not cited as a source here.

Sources

Similar Posts