Wandayi distances Energy Ministry from AI-generated campaign praising tariff decision

Energy Cabinet Secretary Opiyo Wandayi has distanced himself and the Ministry of Energy and Petroleum from an alleged coordinated online campaign that used artificial intelligence (AI) to generate posts praising the government’s decision to withdraw a proposed electricity tariff increase.

The campaign was detailed in a report on coordinated inauthentic behaviour in Kenya, which found that an unidentified actor allegedly used the AI chatbot Claude to generate batches of 50 posts designed to resemble spontaneous grassroots commentary.

A significant portion of the posts praised Wandayi for stopping a planned Kenya Power electricity tariff hike, with hashtags #PowerReliefKE and #PoweringTheNewKenya used to amplify the messaging.

However, investigators did not identify the individual or organisation behind the operation and found no evidence linking it to the government.

Wandayi denies involvement

Responding to the findings, Wandayi said neither he nor the ministry had authorised or participated in the alleged social media campaign.

“My attention has been drawn to some unverified report alleging the creation of some activity on social media that is purported to have been in praise of the actions of the Ministry of Energy and Petroleum regarding electricity tariffs,” Wandayi said.

The CS argued that the existence of anonymous accounts pushing political narratives was not enough to establish a connection with government officials or institutions.

“There are probably millions of anonymous and coordinated accounts on social media pushing different narratives every day. What matters is evidence,” he said.

Wandayi further noted that the report itself did not establish government involvement in the activity.

“The report itself does not establish that either the Cabinet Secretary or the Ministry authorised, directed, or otherwise participated in the alleged activity.”

He said the ministry would continue communicating its work through official channels and would not be distracted by speculation surrounding anonymous online activity.

“We remain focused on communicating the Ministry’s work through credible channels and will not be distracted by speculation around anonymous online activities,” he said.

AI-generated political messaging

The report, identified as GTG-54004: Disrupting a domestic coordinated inauthentic behaviour campaign in Kenya, said the operation was designed to create the appearance of spontaneous public support for the government.

Beyond the electricity tariff messaging, the campaign allegedly promoted narratives suggesting that the United Opposition coalition was breaking apart ahead of the 2027 General Election.

The operation also targeted former deputy president Rigathi Gachagua and retired president Uhuru Kenyatta.

Investigators said the same AI-assisted workflow was separately used to promote Kenyan retail brands through marketing personas identified as “SHANKI” and “Elkins Marketer.”

According to the report, the operation appeared to be entirely domestic, with its pro-administration tone and use of specific hashtags suggesting that it could have been aligned with the ruling coalition.

However, investigators said they were unable to establish the exact organisation responsible.

The report described the messaging pattern as a combination of praise for the government and attacks on the opposition, which it said was consistent with a coordinated electoral communications effort.

How Claude was allegedly used

Investigators found that the template developed for political messaging had previously been used for retail marketing before being adapted to present political content as organic social media commentary.

The actor allegedly used Claude to “humanise” and refine batches of 50 pre-drafted topics and tweets.

The report said the AI tool primarily helped increase the volume of posts and make them appear more authentic, while the underlying political messages had already been determined by the actors behind the operation.

The operation was classified as Category One on the Breakout Scale, meaning its activity remained confined to a network of fake accounts and local influences on a single platform.

Investigators said the campaign failed to reach or influence real people.

Operation taken down

The accounts and organisation associated with the operation were removed following an internal investigation that was triggered by a tip shared by OpenAI regarding recidivist activity on its platform.

The investigation also led to the development of detection methods based on the operation’s behavioural patterns, according to the report.