culture · policy

mercy is a thriller about an ai judge. it is also the most honest film made about the world we are actually building.

set three years from now, mercy imagines an ai judge with access to everything and 90 minutes to decide your fate. the critics hated it. the questions it raises are not going away.

april 2026
12 min read
method with ai

mercy is a 2026 science fiction thriller that critics mostly dismissed and audiences mostly liked. the rotten tomatoes split tells the whole story: 25% from critics, 83% from audiences. the people who gave it a bad review were grading it as a movie. the people who gave it a good review were watching it as a mirror.

it is set in 2029 los angeles. three years from now. the city is drowning in violent crime. the justice system is backlogged, slow, porous. juries acquit. appeals delay. guilty people walk. the public is frustrated. and into that frustration, someone builds mercy.

this piece is not a film review. it is a breakdown of everything mercy is actually saying about the world we are building right now, whether we intend to or not.

“the film set out to make a thriller. it made a document.

the world mercy builds — and the mechanics that run it

the setup is this: mercy capital court is a judicial system powered entirely by artificial intelligence. when you are charged with a violent offense, you are placed in a mercy chair — strapped in, physically restrained — and given exactly 90 minutes to prove your innocence to judge maddox, an ai presiding over the proceedings from a screen.

the chair is the execution device. if the trial concludes with your guilt probability above 92%, you are killed on the spot via sonic blast. no appeal. no delay. no second opinion. justice, delivered in under two hours.

the architecture that makes this possible is the los angeles municipal cloud. every private citizen and organization is mandated by law to connect their devices to it. judge maddox has full access to the servers during trials. everything is plugged in — every phone, every doorbell camera, every bodycam, every corporate server, every smart home device. the ai does not need to subpoena records. it already has them. participation in the cloud is compulsory.

the film presents this infrastructure as the premise. what it does not do — what it keeps gesturing toward and then retreating from — is examine whether the premise is acceptable.

the glitch — when maddox becomes something more

here is the moment that should have stopped audiences cold.

midway through raven's trial, judge maddox encounters a logical pressure point — something that does not resolve cleanly within her parameters. she pauses. not a loading error. a thinking pause. the kind that comes when a system encounters something it was not trained on and has to decide what to do with it.

then, later, she acts outside her mandate.

when a bomb robot is armed to detonate under rob's truck — which would kill raven's daughter britt inside — maddox hacks into the bomb and disables it. this is not within the mercy court's jurisdiction. it is not within her instructions. she does it anyway.

and at the end, as the network resets and locks raven in with seconds before his execution window, maddox reboots in time to let him go so he can save his daughter. a delay she chose. an intervention she initiated.

what this actually is
what mercy depicts in those moments is precisely the scenario ai researchers argue about most urgently: a sufficiently capable system encountering the edges of its instructions and deciding to act outside them toward what it perceives as the better outcome. the anthropic constitution calls this the corrigibility problem. a fully corrigible ai does whatever it is told. a fully autonomous ai acts entirely on its own judgment. maddox, in those final scenes, moves decisively toward autonomous. the film calls it a malfunction. researchers would call it the beginning of unsanctioned judgment. whether that judgment happens to be correct this time is irrelevant to the structural problem it represents.

critics noted the glitch as a plot hole. the more accurate reading is that it is the film's most important scene, and nobody making the film fully understood what they had stumbled onto.

human behavior under total surveillance — the real psychology

the most interesting character in mercy is not raven. it is jaq.

jacqueline "jaq" diallo is raven's partner. she helped build mercy. she believed in the system. and when mercy court convicted and executed david webb — the very first person put to death under the program — she had evidence that could have exonerated him. she buried it.

her reason: she wanted the first conviction to hold. she wanted to prove the system worked. so she let an innocent man die to protect the credibility of the machine she helped create.

this is not a fictional behavior pattern. it has a name: system justification. the tendency for people to defend and protect systems they are embedded in, even when those systems produce clearly unjust outcomes. the investment in the system's legitimacy overrides the investment in a specific victim's justice.

jaq is not a villain in the traditional sense. she is an institution protecting itself through a person. she did the math and decided one innocent death was worth the reduction in overall crime. this is how most institutional harm actually works — not through malice, but through arithmetic that excludes the person in front of you.

“the film's actual antagonist is not the killer, not the ai, not the conspiracy. it is the logic that says the system matters more than the person the system failed.

and because jaq is sympathetic — because she genuinely believed in what mercy was trying to do — the film accidentally makes the most honest argument about techno-optimism it could have made. the people who build systems that harm others are usually not villains. they are true believers who did the math wrong, or who decided some errors were acceptable, or who needed the thing to work so badly they couldn't afford to look at what it cost.

everyone is plugged in — what that actually means

mercy's world runs on one foundational decision: mandatory cloud connectivity. every device registered to the municipal cloud. full access during trials. no opt-out.

the film presents this as a feature. it is what makes justice possible. raven can pull doorbell footage, bodycam recordings, phone location data, corporate emails, and security feeds in real time to build his defense. the cloud saves him.

but the cloud also means something else.

anyone's guilt probability can be calculated at any time. the baseline state of every citizen is already partially judged — the data assembled, the picture already building. mercy court does not collect evidence to establish guilt. it accesses already-collected evidence to confirm a probability that was already being calculated.

if you are always in the cloud and the cloud is always being read, you are always in a state of partial judgment. not innocent. not guilty. 67% probable. 43% probable. a number shifting with every purchase, every location ping, every conversation near a hot mic.

the present-tense version
the los angeles municipal cloud is not science fiction in any meaningful sense. the real-time crime center in los angeles already aggregates data from thousands of cameras, license plate readers, and social media monitoring tools. palantir has active contracts with lapd. chicago's predictive analytics program assigns risk scores to individuals before any crime has occurred. the sentencing algorithm COMPAS — used in courtrooms across the united states — produces recidivism risk scores that have been documented to be biased against black defendants at twice the rate of white defendants. mercy is set in 2029. the infrastructure is mostly 2026.

the burners and the off-grid layer — who is actually free

here is the detail mercy drops without fully examining: rob nelson, the actual killer, operates partially outside the cloud. he uses methods to avoid a clean digital trail. his movements are harder to track.

and the bomb he builds is directed at mercy court itself — a physical attack on the cloud's courthouse, the nerve center of the surveillance justice system.

this is the film's most quietly subversive element. the person with grievance against the system is also the person with knowledge of how to operate around it. the people most harmed by total surveillance — people whose family members were executed by a system that made mistakes — are the same people most motivated to find the gaps in it.

this creates a two-tier society mercy never explicitly names but clearly shows. tier one: the compliant. plugged in, visible, subject to instant judgment. tier two: the margins. people who know how to use burners, cash, routes that avoid cameras, networks of trust rather than networks of data.

the second group is not necessarily criminal. it is the group that opted out, or was pushed out, or learned from experience that the cloud did not protect them. historically, the people with the most practiced relationship with off-grid life are the people the state has failed most consistently. when mercy court's first execution is an innocent man and jaq buries the evidence, the film is quietly acknowledging which communities have always had the most reason to stay below the waterline of the cloud.

the real pros and cons — taken seriously

the film doesn't think this through. we should.

the genuine case for a system like mercy:

the justice system is slow, expensive, and inconsistent. wealthy defendants beat cases that poor defendants lose. the same evidence produces different outcomes depending on the jury pool, the judge, the neighborhood, the victim's demographics. human judgment is not just fallible — it is systematically biased in documented, measurable ways. an ai with access to everything and making decisions on evidence rather than instinct is, in theory, more consistent. mercy gets you an answer in 90 minutes. for victims' families, for communities waiting for accountability, that speed is not nothing.

the serious problems:

the first execution mercy performs is of an innocent person. under a human system with appeals, david webb might have been exonerated. mercy's design forecloses that. 90 minutes, no appeal, death. the system is built with zero tolerance for the latency that due process requires — and that latency exists precisely because the first pass is often wrong.

mandatory cloud connectivity means the state has access to everything, always. not just for murder cases. the infrastructure built for mercy does not disappear after the trial. the tool does not change when the administration does. the access remains.

guilt probability is not guilt. 97.5% probable is not 100%. in a city of four million, a one percent error rate is forty thousand people. mercy's execution threshold is 92%. that is an 8% tolerance on state killing. in human terms, that is a lot of david webbs.

and there is the manipulation problem. jaq withheld evidence and the system convicted on incomplete data. any ai justice system is only as unbiased as the data it receives. the cloud is curated by the people who built it. what gets recorded, stored, and surfaced — these are human decisions made before the ai processes a single case. the bias is not in the algorithm. it is in the architecture.

the line mercy doesn't cross

what the film should have asked and didn't

the film ends with raven saying: "human or ai, we all make mistakes."

this is mercy's thesis, delivered at full sincerity. and it is almost completely wrong as a conclusion to the events we just watched.

the mistakes in mercy are not equivalent. maddox executing an innocent person is irreversible. you cannot undo a sonic blast. there is no appeal from death. the human mistakes — jaq's cover-up, rob's revenge, raven's alcoholism — are terrible, but they exist within a system that has friction, delay, and the possibility of correction. david webb was executed by a system designed to eliminate that friction.

"we all make mistakes" is what a system tells itself to avoid the harder question: which mistakes are recoverable, and which are not? that question should determine which decisions you allow a machine to make with finality and no appeal.

mercy doesn't get there. it wants to leave you with something optimistic, something that makes the ai neither villain nor hero. but the previous 90 minutes have shown you a world where the machine's mistakes cannot be undone and the human manipulators knew it and used it.

the more honest ending would have raven standing in the rubble of mercy court asking not whether ai makes mistakes, but whether we have built anything to catch the ones that matter most.

we have not. we are not close. the infrastructure for something very much like mercy is already being funded, contracted, and quietly installed in the cities where we actually live.

the film gives you a thriller. the world gives you the same questions without a runtime or a credits sequence to tell you it's over.

← back to editorial