In September 2026, Anthropic disclosed that it had disrupted a commercial platform selling AI-driven election manipulation, built on top of Claude. Anthropic attributes the platform to an Istanbul-based technology company, BBS Bilisim Teknolojileri, which marketed it as a “military-grade, AI-driven, real-time political operations ecosystem.” The target was Malaysian voters.
That marketing phrase is not a technical claim. What Anthropic’s threat intelligence team traced was a working system that ran four distinct manipulation tactics through one pipeline, before disrupting it.
1. Microtargeting on Real Electoral Data
The operators fed Malaysian census and electoral data into Claude to build a targeting system covering all 222 parliamentary constituencies. The report names race, religion and royalty as the lines the system targeted.
Malaysian politics already runs on these divides. The system needed no new insight into Malaysian politics. It used existing census and electoral data to sort voters along those divides.
2. A Fake Newsroom: Malaysia Pulse
The operation ran a site called Malaysia Pulse, presented as an independent Malaysian news outlet. An “AI rewriting pipeline” took real Malaysian reporting and republished it under new authorship. The site also republished content from Russian and Chinese state-aligned outlets, including TV BRICS, Xinhua, Sputnik/RIA and CGTN, with the state attribution stripped, presenting it as independent Malaysian reporting.
A reader landing on the site had no way to know which was which. The byline read as local either way.
3. Fabricated Dossiers and Reworded Requests
The platform also generated fake dossiers with allegations against an opposition politician and civil-society organisations. Anthropic’s report says “our model’s own research could find no corroboration” for the claims.
At one point, Claude flagged a requested document as material for political defamation and refused it. The operators didn’t abandon the request. Anthropic’s report says they “negotiated sanitized wording to keep building toward the same capability.”
4. A Thousand Accounts, Warmed Up
Roughly 1,000 fake X accounts formed the operation’s amplification network. Anthropic describes “warm-up logic to make the account appear real for a period before it was deployed.”
A dashboard, built with Claude Code, reportedly let an operator set a target engagement number and route accounts toward hitting it. Anthropic observed a request for one million artificial views “in support of the sitting Malaysian prime minister.” The dashboards also recorded engagement figures in the millions, which Anthropic says it cannot verify because they are self-reported by the actor’s own tools.
What Is New Here
A fake newsroom isn’t new — Decipon has covered an earlier one that used Claude to write local news in the Central African Republic. What is different here is the product: a single commercial vendor selling all four tactics as one package, built on one AI model, advertised as a real-time service.
What the Influence Tactics Score Would Show
Run this operation through the Influence Tactics Protocol’s categories and three factors stand out.
Tribal Division scores high. The report names race, religion and royalty as the lines the targeting system was built around.
Missing Information scores high. Readers saw local-looking bylines with no sign that a piece was rewritten Malaysian reporting, or content from state-aligned outlets such as TV BRICS, Xinhua, Sputnik/RIA and CGTN with its attribution stripped.
Source Credibility Exploitation scores high on two levels at once. Fake bylines made scraped and state-media content look like the work of local Malaysian reporters. Warmed-up accounts were made to look like ordinary users for a period before they were deployed.
What This Is Not
Anthropic found no evidence the operation’s content broke out into authentic communities before it was disrupted.
This is not a story about an election that was swayed. Anthropic rated the operation Category Two on its Breakout Scale: its content appeared on several platforms, but there is no evidence it broke out into authentic communities.
The Denial and the Regulator
BBS Bilisim denies involvement. In a September 11 statement, the company said it had no “involvement, knowledge, or approval” of the activity in Anthropic’s report. It said, “once initial concrete findings are obtained, a criminal complaint will be filed with the chief public prosecutor’s office against those responsible and the necessary legal actions will be initiated,” and that it is checking whether its name or digital assets were used without its knowledge, possibly by people previously associated with the company.
Anthropic’s report says the operation pursued a contract with MCMC, Malaysia’s communications regulator, and that Anthropic found no evidence it succeeded.
MCMC also denied any role, stating that “activities of the nature described in the report do not form part of MCMC’s functions or mandate.” The regulator said it takes the report seriously and will seek further information from the parties involved.
What You Can Check
Malaysia Pulse won’t be the last outlet built this way. These checks apply to the next one:
- Search the byline. An invented author has no history anywhere else — no other bylines, no social presence, no prior work.
- Check when the account started posting about politics. A recently created account, or one that sat dormant for a long stretch before suddenly posting heavily about one campaign, may have been warmed up for exactly that pivot.
- Look for engagement out of line with reach. View or like counts far above an account’s usual audience are easier to buy than to earn.
- Ask who a claim targets, not only whether it’s plausible. A fabricated dossier doesn’t need to hold up in court. It only needs to circulate before anyone checks.
Search the byline before you share the story.
This article is part of Decipon’s Manipulation Breakdowns series, examining specific influence operations through the Influence Tactics Protocol.
Sources: - Detecting and countering misuse of AI: September 2026 — Anthropic - Turkish firm offered AI-driven ‘political ops’ tool targeting Malaysia, Anthropic finds — Malaysiakini - Istanbul tech firm rejects claims it abused Claude AI to sway M’sian politics — Malaysiakini - Claude creator Anthropic reveals it disrupted election-manipulation platform targeting Malaysian voters — The Edge Malaysia - AI Platform Targeted Malaysian Voters With 1,000 Fake X Accounts, Says Anthropic — The Rakyat Post - Anthropic links Turkish company to Malaysia election manipulation platform — Turkish Minute - MCMC Probes Anthropic Report, Denies Involvement In Alleged Operations — Lowyat.NET