PRISONBREAK: when even your liberation is fabricated

Let us start with a confession: the propaganda of your enemy's enemy is the most digestible propaganda on earth. If you despise a regime, any content that hits that regime strokes your priors. And that is exactly where the professionals are waiting.

The Citizen Lab at the University of Toronto — an outfit with a long record of documenting spyware and influence operations — documented a network codenamed "PRISONBREAK": Israeli-backed, pushing AI-generated anti-government propaganda at Iranian users. Savor the codename. You were meant to break out of the prison of the official narrative — straight into a narrative drafted by a different staff office.

Cell by cell

In the deepfake case files hub we documented how networks linked to the Iranian government produced a flood of fabrications, with Moscow and Beijing handling distribution. PRISONBREAK is the mirror piece of the same puzzle, with the firing direction reversed. One factory wanted you to believe the regime was winning; the other wanted you to believe it would collapse by morning. Neither asked your opinion. Both wanted only your behavior.

There is a subtlety here worth protecting: the documentation of PRISONBREAK does not mean "so Tehran was right after all." A state running its own fabrication factory cannot be acquitted by the discovery of its rival's factory. Two lie factories do not cancel each other out; they merely double the smog.

Why the Iranian audience?

Because it was the most valuable terrain of the war. How the population inside Iran behaved — whether it hoped or despaired, stayed home or took to the streets, laughed at the official story or believed it — carried operational value for both belligerents. And once your behavior acquires military value, your feed becomes a military target. That simple, that shameless.

The irony is exquisite: this target audience had to switch on a VPN and climb a censorship wall just to reach the shooting range. The state built the wall so its people would see nothing but the official narrative; on the far side of the wall, another staff office sat waiting with a narrative pre-written for those same people. We describe that wall in VPN nation. The prisoner escaped one wing and arrived in the next.

Detection is getting harder, not easier

The previous generation of psy-ops could be unmasked with diligence: broken Persian, stock photos, accounts three days old. AI has optimized those human errors away. Today's material has the right dialect, the familiar pain, even the well-timed joke. And when users asked a chatbot whether a clip was real, they got the outcome we documented in When AI verifies AI: the verification machine authenticates the other machine's forgery.

What remains is the old, merciless question: what does this content want from me? If the answer appears to be "nothing, it just informs" — double your suspicion. Disinterested information is the most expensive commodity in the market, and nobody hands it out free.

Closing analysis: the enemy of your enemy is not your friend

Diaspora audiences long ago learned to read state media with suspicion — correctly. PRISONBREAK is the reminder that the suspicion cannot run one way. Whoever fabricates against your oppressor is not fabricating for you; they have plans for you. Recall the two broadcasting machines that have fought over the same Iranian viewer for decades — the story is in The Persian media mirror — then imagine that same fight with an AI engine bolted on.

Your trust is capital. Both sides have set up departments for its expropriation. Spend it only against an invoice.

Back to the archive

This piece is one case in our information-war dossier; the full archive and how to read it live at the deepfake case files hub.