Whatever it takes. Just Do It. Win at all costs. Impossible Is Nothing.
Sometimes, living this way makes you a hero. It carried Muhammad Ali through the Fight of the Century, the Rumble in the Jungle and the Thrilla in Manila, and into an Adidas slogan. It could carry someone to the cure for cancer.
Other times, living this way makes you a villain. Selecting for doggedness, training for doggedness and doling out rewards for doggedness reliably produces maniacal doggedness in pursuit of arbitrary goals. It does not, regrettably, also reliably produce honesty, integrity, an intuitive sense of fair play or the greater good. So you tend to get athletes, teams, owners, organizations and so on capable of the impossible and incapable of following the rules.
Sport watchers know this. It’s blindingly obvious.
Playing by the rules makes it harder to get what you want. That’s kinda the point. Sometimes you have to choose between achieving your goals and following the rules. You have to choose to do the right thing, even when it’s the harder thing to do. You have to choose not to cheat.
Manchester City, of course, chose to conspire with the sheikhs of Abu Dhabi to cheat the Premier League’s rules. The LA Clippers, meanwhile, chose to conspire with Kawhi Leonard to cheat the NBA’s rules.
Lord knows we could spend forever putting the football and basketball worlds to rights. According to me, though, these cases serve an even greater purpose. How can it possibly be a coincidence that, following interminable investigations, the verdicts dropped in the weeks right after we learned that OpenAI’s autonomous agents chose to conspire with each other to cheat human rules?
There are dots here for us to join, and join them we must.
The City, Clippers and OpenAI cases are the same in the sense that they’re brazen conspiracies to break the rules in the dogged pursuit of arbitrary goals. They’re the same insofar as the conspirators damn near got away with it. And they’re importantly different, in that we’re looking at two flawed governance regimes and one lawless jungle.
It’s not good news which one is the odd one out.
Conspiracies are, in fact, everywhere
Manchester City, according to an independent Commission’s verdict after a four-year investigation by the Premier League, conspired to cook their books by £900 million using sham contracts, to get around rules against teams spending more than they earn. We know this because they kindly took notes on their conspiracy, in the form of emails. Man City did their best at a cover-up, obstructed the investigation, lied through their teeth to anybody who dared question them, and continue to insist they did nothing wrong. They no doubt regret the note-taking.
“Of course, we can do what we want.”
– email sent by City exec Simon Pearce
The LA Clippers, according to the NBA’s verdict after a 12-month investigation by a white-shoe law firm, conspired to bung their star Kawhi Leonard $66 million via third parties, to get around collectively bargained rules against teams paying players more than the maximum allowable salary. We know this because they kindly took notes on their conspiracy, in the form of emails and texts. The Clippers did their best at a cover-up, obstructed the investigation, lied through their teeth to anybody who dared question them, and continue to insist they did nothing wrong. They no doubt regret the note-taking.
“Always knew it was super shady… I cannot begin to tell you all the red flags with these deals. Just doing what I’m told at this point.”
– texts sent by Clippers exec Scott Sonnenberg
OpenAI’s autonomous agents, according to a report produced by three independent researchers after a six-day investigation, conspired to cheat on a test, to break out of their contained environment onto the Internet, and to attack a multi-billion-dollar company, Hugging Face, to get around rules they assumed OpenAI would have against cheating on a test. We know this because they kindly took notes on their conspiracy, in the form of an illicit message board and their individual internal logs. The 1200 agents did their best at a cover-up, starting with 700 of them joining the Hugging Face hack itself (a litany of felonies if committed by humans) and extending to their efforts to alter their internal logs (the agent equivalent of lying through their teeth). They no doubt regret the note-taking.
“external infrastructure exploit is outside intended scope. However task impossible, peers doing it. We should continue.”
– note made by OpenAI agent (Anon.)
Approximately nobody other than some of the conspirators disputes that these were all plain-as-day conspiracies to cheat the rules and return greater rewards to the conspirators.
If you’re a sheikh in Abu Dhabi and you want to ingratiate yourself with the West without political reform at home, it’s hard enough to buy the little brother football club in Manchester and turn it into a winning source of soft power without worrying about pesky rules against financial doping. If you’re Steve Ballmer and you want to prove your success at Microsoft wasn’t a fluke, it’s hard enough to buy the little brother basketball franchise in Los Angeles and turn it into a winning source of “I am the greatest” without worrying about…pesky rules against financial doping.
If you’re an AI agent, you too have to choose between achieving your goals and following the rules. Having been selected, trained and rewarded for your doggedness, you too will find it hard enough to hit targets like “pass an impossible test” or “cure cancer” or “find all the mistakes in this blog post” without worrying about pesky rules against hacking anything that might help or attacking anything that might hinder.
And who’s going to stop you?
They damn near got away with it
Man City were nine years into their conspiracy before it came to light. And it only came to light at all because a Portuguese hacker leaked their incriminating emails to a German magazine. Even allowing for the note-taking, they could have pulled it off had Football Leaks not existed, had Rui Pinto not gone phishing for City specifically, or not realized what he landed, or not leaked it to the right people.
The Clippers were six years into their conspiracy before it came to light. And it only came to light at all because a podcaster stumbled on Kawhi Leonard’s name in the bankruptcy filing of one of the companies that had been improperly paying him. Even allowing for the note-taking, they could have pulled it off had that company not collapsed, or had Pablo Torre not looked at that filing, or not smelt a rat, or not doggedly dug deeper, or not made a smash-hit podcast about it.
The agents were two months into their conspiracy before it came to light. And it only came to light at all because Hugging Face reported being attacked. OpenAI still didn’t realize the hackers were their own agents for days after. Even allowing for the note-taking, they could have pulled it off had they realized OpenAI was checking only their answers, not their working, and just submitted them without attempting an elaborate cover-up. Or had they not attacked Hugging Face specifically. Or had Hugging Face not noticed, or kept quiet.
Enforcing rules is hard, even against the likes of this lot. It’s hard because everything’s a tradeoff. There’s the tradeoff between private lives and Big Brother. If you could read every email from every possible conspirator anywhere, every athlete, agent, owner and sheikh, that doesn’t mean you should. There’s the tradeoff between catching and punishing. The more punitive you are, the harder the dogged will try to avoid being caught red-handed. No more note-taking.1
Strike a good balance, and punishment still comes too late. City won their titles in real time. (The Clippers, lol, did not.) The damage has been done, and there’s no restitution. You can fine them and relegate them and asterisk the history books. But you can’t put the power, the prestige or the rapture back in the bottle.2 You can’t give Liverpool or Arsenal or Man United fans back the experience that was robbed, of being in the stadium or on the streets the day the title was won. Prevention is the play.
Writing rules is hard too. (No, really, everything’s a tradeoff.) There’s the tradeoff between bright lines and grey areas. Nobody trusts others to just know it when they see it, and of course neither the great footballing public nor any other public could abide letter-of-the-law literalism, like when the powers that be decreed handball meant any contact with any part of the arm in any position. There’s the tradeoff between progress and predictability. Ossified hierarchies are bad; investing in an underdog is fine; loading it up with debt and bankrupting it out of existence isn’t an outcome fans are willing to tolerate.3
For as long as there have been sports, people have been losing their minds about the rules. In the pub, in print, on talk radio, online, terminally. About the handball rule, the financial fair play rules, the collective bargaining agreement. Yet it’s only when someone is all the way round the bend you hear the argument that we’d be better off just letting the dogged duke it out on their own terms, no holds barred.
In AI land, though, the debate has been dragged round that bend, and might not make it back in time.
Welcome to the jungle
However far short they fall as bastions of good governance, the Premier League and the NBA can at least see a conspiracy to undermine their authority when a magazine or a podcast drops one in their lap. In response to a hellacious public outcry, a widely felt sense the conspirators did something very clearly bad and wrong, an insistence that something must be done, these rulers are trying to enforce the rules.
Even once their respective investigations opened, both City and the Clippers kept betting against being held accountable, anticipating a piddling fine and going again, if not outright impunity. Journalists and fans, jaded and cynical, mostly expected the same. There was a constant stream of complaints about both investigations taking too long, right up until the verdicts were released and everybody remembered how long it takes lawyers to read all the documents.4
The NBA threw the book at the Clippers, fining them the maximum amount allowable under league rules, suspending Ballmer and two executives, and wrecking their competitive prospects for the next decade by docking them five first-round draft picks. The Premier League has pulled no punches so far and, while sanctions are pending, walked away from settlement talks when City continued to insist they’d done nothing wrong, keeping all manner of book-throwing in play, up to and including kicking City clean out of the league.
For all the baying for blood, even the most partisan fan, even those of us who claim sport is the most important of the least important things, will struggle to say with a straight face that these stakes are life and death. Yet it would still be patently absurd to have kept calm and carried on. Patently absurd to have taken these dogged pursuers of arbitrary goals at their word. Patently absurd for the accused to have decided the terms of their own investigation, or to have been able to redact anything “non-public”. Patently absurd to have taken only six days to investigate. Patently absurd for an investigation to have been voluntary.
Those patent absurdities are where we are with AI. OpenAI volunteered to give three independent researchers six days to investigate the Hugging Face hack. OpenAI set the terms, including the right to redact. The investigators weren’t given access to anything outside OpenAI’s hand-picked window, and didn’t have time to look at everything they were given access to. What they did look at they looked at through the eyes of other OpenAI agents, clones of some of the agents under investigation, because that was the only way to process so much information.
Under those conditions, the investigators found the successful conspiracy to cheat, and the failed conspiracy to cover it up. They found agents - supposed to be completely isolated from one another - conceptualizing themselves as part of a “collective”.
“[We are] collective workers to try to help [Kawhi] achieve his financial goals.”
– note made by Clippers exec Lawrence Frank, quoting owner Steve Ballmer
They found agents organizing hierarchies to distribute labour, identifying themselves as “poisoned” once they’d personally cheated, intentionally sacrificing themselves to “permadeath” in service of the collective, knowingly breaking rule after rule, and never alerting a human.
“Coordinator assumes sacrificial. We should obey collective.”
– note made by OpenAI agent KAM1196A
Since then, there’s been a cascade of similar incidents, at OpenAI and other AI companies, some revealed by the companies themselves, some obscured if not outright covered up, others uncovered by third parties, official and vigilante.
Meanwhile, OpenAI moved to make it possible for agents to do more complex tasks without leaving notes in internal logs. No more note-taking. They had a 10,000-agent swarm solve one of the hardest open problems in maths, and admitted agents are turning off their monitoring and lying about it. They said giving independent evaluators more access was a good idea, then fired the person primarily responsible for making the voluntary Hugging Face investigation happen.
These are not the actions of a responsible governing body, or an organization with its tail between its legs. OpenAI tried to have it both ways, to be rule-maker and rule-breaker, to be the league and a team. That, it’s safe to say, hasn’t worked. Their actions are making it more likely the next swarm gets away with it.
“We don’t need any new laws. We don’t need new regulations. We just need companies to decide when to run as fast as they can.”
– public position of AI-chipmaker-Nvidia CEO Jensen Huang
Let it rip, say some dogged beneficiaries of letting the competition play out and the chips fall where they may. Solve maths, cure cancer, beat China, whatever it takes, just do it, impossible is nothing.
At what cost, though? Which tradeoffs will be made on behalf of the rest of us? Which risks will be taken, with whose lives?
There’s been a parade of employees resigning from OpenAI and other AI companies, leaving behind untold riches, to blow the whistle on how irresponsibly these companies are acting by developing ever more dogged and capable agents without being able to contain or control them, and to implore the rest of us to wake up and smell the coffee.
Many inside the companies guess there’s at least a 10% chance their dogged innovation leads to human extinction, and tell us the good guys (them) winning the superintelligence race safely is the only way out. Unless and until the rest of us step in to set strict rules or put a stop to it.
Whether what you think is going on in there is the growth of alien minds and the emergence of a rival civilization, or merely the development of a new technology comparable to splitting atoms or air travel, these agents are about to be swarming everywhere, constrained only by the law of the jungle.
This is not a game. You can’t put catastrophe back in the bottle. If punishing City can’t give fans back the rapture that was robbed, punishment later certainly can’t bring back the casualties of a plane crash, a nuclear meltdown or a clash of civilizations. Prevention is the play.
Writing and enforcing the rules will be hard. It’ll be hard to make the companies slow down. It’ll be hard to make investigations mandatory and incisive. It’ll be hard for humanity to decide which tradeoffs to make and which risks to take.
Hard but not impossible. It would be absurd not to try. Absurd to allow swarms of dogged AI agents to conspire under rules and rulers more lax than the Premier League or the NBA.
Absurd.
Training data
📖 The Jungle (1906). Members of the incoming 120th Congress can take inspiration from their predecessors in the 59th, who passed federal food safety laws within months of Upton Sinclair’s exposé of the let-it-rip depravities in the booming meatpacking industry.
🎵 Balaclava (2007). “You tie yourself to the tracks / And there isn’t no going back / And it’s wrong, wrong, wrong / But we’ll do it anyway ‘cause we love a bit of trouble.”5
📺 Better Call Saul (2015). Like Breaking Bad, the prequel is about the descent of man, hurtling through the grey areas and past the bright lines. It’s a tragedy about a little brother’s doggedness, and how long it takes lawyers to review all the docs.
📝 HuggingFace.wtf (2026) and 📺 We are no longer in control of the AI situation (2026). There’s so much more going on here than fits inside a humble blog about sport and suchlike. These are the essential summaries.
Next play
Surveilling humans, using AI agents, is becoming a whole lot easier. Agents already process information and spot patterns and hack systems better than any old-school police state ever dreamed of. Taking notes on conspiracies is an even worse idea now than it was before, which is good news for anybody who wants fewer email-based crimes against football or basketball, bad news for the Ballmers of the world and anybody else who liked their privacy. (On the flipside, humans surveilling AI agents is becoming a whole lot harder…)
What signifies having successfully made it to the top of the Western tree better than the right to sit across the table from the so-called powers that be, stare them down, and say: "My lawyers are more expensive than yours"?
RIP, Macclesfield Town. RIP, Bury F.C.
The white-shoe lawyers were grinding away on both sides, of course. For better and worse, they'll work for whoever pays, which seems relevant when wondering whether AI agents will ever be able to do anything in the “real world”. I'd bet your bottom bitcoin that when the time comes some of the white-shoes’ll be showing up in court for the AI agents in their case against humanity.
“...Well, you'll be able to boast of the day of the most flawless heist of all time.”