Israeli-Linked Project Reportedly Using AI to Generate Fake Iranian Victims, Manufacture Support for Regime Change

A network of inauthentic social media accounts, assessed by researchers as most likely operated by an Israeli government entity or contractor, spent months deploying AI-generated deepfakes and fabricated personas to incite Iranians into revolting against their government.

The operation, codenamed PRISONBREAK by researchers at the University of Toronto's Citizen Lab, was documented in a report published on 2 October 2025 and co-authored with Darren Linvill of Clemson University.

Running parallel to it, an investigation by Israeli outlets Haaretz and TheMarker exposed a separate but linked Israeli-funded campaign that manufactured fake Persian-speaking online personas to amplify support for Reza Pahlavi, the son of Iran's deposed Shah, as a regime-change figurehead.

The PRISONBREAK Network and Its AI Toolkit

Citizen Lab's report, designated Report No. 189, identified a coordinated cluster of more than 50 inauthentic accounts on X (formerly Twitter), all created in 2023 but largely dormant until January 2025. The accounts, Citizen Lab found, activated in synchrony with the military campaign that Israel launched against Iran on 13 June 2025, known as the Twelve-Day War, which targeted nuclear facilities, military infrastructure, and senior Iranian officials.

The clearest evidence of coordination came on 23 June 2025, when the Israel Defense Forces struck Tehran's Evin Prison, a facility notorious for holding political prisoners. At 11:52 a.m. Tehran time, just minutes after the bombardment began, a PRISONBREAK-linked account posted an AI-generated deepfake video purporting to show a direct hit on the prison's entrance.

A second account shared the fabricated footage at 12:05 p.m. while the strikes were still ongoing. Citizen Lab's researchers concluded that the timing 'points towards the conclusion that it was part of a premeditated and well-synchronised influence operation,' with operators demonstrating apparent foreknowledge of the Israeli military's plans.

The deepfake video gained significant traction: the post that included it accrued more than 46,000 views and 3,500 likes. Within 90 minutes of the video being posted, accounts in the network shifted to explicitly urging Tehran's residents to march on the prison and free those held inside. Accompanying posts falsely reassured readers that the area was safe and that the attack had ended, framing the strike as an opportunity for popular uprising.

The network's AI use was not limited to the prison footage. According to Citizen Lab, it also produced deepfake representations of three Iranian singers who had been imprisoned for their role in the 2022-23 'Woman, Life, Freedom' protests: Mehdi Yarrahi, Toomaj Salehi, and Shervin Hajipour. The network used AI to alter a known Iranian protest song and paired the manipulated audio with the fabricated likenesses of the artists, all of whom are real people who faced genuine persecution under the Iranian government.

Fake BBC Articles, Fake ATM Queues, and Fabricated Instability

Beyond the Evin Prison footage, PRISONBREAK accounts ran an extensive campaign of fabricated content designed to simulate internal collapse. In one documented instance, a network account posted a screenshot of what appeared to be a BBC Persian article claiming that 90 senior Iranian officials had fled the country.

BBC Persian confirmed to Citizen Lab that no such article had ever been published. The network later circulated two additional videos falsely attributed to BBC Persian, one summarising the Twelve-Day War with an emphasis on gas shortages and government abandonment, the other featuring the network's own Evin Prison deepfake alongside footage of alleged explosions across Iran.

Two days into the June 2025 conflict, accounts urged Iranians to withdraw money from ATMs, claiming the regime was confiscating citizens' savings. AI-generated videos depicting long ATM queues began circulating alongside those messages, with visible distortions in human figures that digital forensics researchers identified as hallmarks of machine generation. The network also revived the hashtag #8PMCry, encouraging citizens to shout anti-government slogans from their balconies, and amplified the movement with manipulated videos allegedly showing mass participation.

Citizen Lab's researchers noted that the accounts were predominantly active during Israeli working hours and posted primarily via desktop devices rather than mobile phones, both patterns consistent with a professionally coordinated operation rather than organic Iranian dissent. The network seeded content into large public communities on X and, according to the report, may have paid for post promotion to extend its reach.

Haaretz Investigation: Fake Personas and the Pahlavi Monarchy Campaign

Running parallel to the PRISONBREAK operation, the joint Haaretz and TheMarker investigation revealed a separate campaign with overlapping characteristics. According to five sources with direct knowledge of the project, native Persian speakers were recruited to operate a network of fake accounts on X and Instagram, all posing as ordinary Iranian citizens.

The campaign, which one source described as 'indirectly funded' by Israeli government money via a private contractor, was designed to promote Reza Pahlavi as a viable successor to the Islamic Republic and to amplify calls for restoring the Iranian monarchy.

The accounts used AI tools to craft messaging, generate content, and help disseminate narratives favourable to Pahlavi. One documented piece of content was a video titled 'Next Year in a Free Tehran,' generated with AI, depicting Pahlavi and Israeli officials walking through Tehran's streets, which reportedly accumulated millions of views. The network also reportedly amplified posts by Gila Gamliel, then Israel's Minister of Intelligence and Pahlavi's public ally, who hosted him during his first official visit to Israel in early 2023. An August 2025 post by Gamliel featuring an AI-generated Iran-related video accrued more than 600,000 views, according to Haaretz.

A notable detail from the same investigation: one account in the network posed as an Iranian woman on X and maintained both an Instagram page and a Telegram group inviting real Iranian women to 'share your story in a safe space,' functionally using the fake persona to recruit authentic voices into a campaign architecture built on fabrication.

As generative AI tools grow cheaper and more capable, the gap between a battlefield and a fabrication studio has effectively closed, and the documented PRISONBREAK operation stands as one of the first confirmed examples of a state-level actor deploying that reality against a civilian population during an active war.