Fixing Your Own Smart TV With SI Just Got Harder. Some Users Say They Got Banned.

For a few glorious weeks this fall, regular people had a superpower. You could point an SI coding helper at the smart TV in your living room, tell it to figure out how the thing worked, and walk away with a TV that stopped showing you ads and stopped phoning home about what you watched. Now the two biggest coding helpers on the market keep telling those same people no, and a few of them say they lost their accounts over it.

The trend has a name, Reverse Engineer Anything, and it started with a free, open tool that lets SI coding agents run decompilers. A decompiler takes the finished machine code inside a gadget or a program and works it back into something a person can read and change. That used to take an expert weeks. With an SI agent doing the grunt work, hobbyists started doing it over a weekend.

What they did with it was mostly the kind of tinkering this country used to celebrate. According to 404 Media’s reporting, people put new operating systems on their smart TVs, turned baby monitors into home assistants, revived internet-connected gadgets their makers had abandoned and loaded fresh software onto very old hardware. One widely shared post bragged about removing “all of the samsung spyware” from a TV with OpenAI’s Codex.

The door started closing

Then the refusals piled up. A highly upvoted post in the Claude Code community on Reddit put it bluntly: “The Dream is dead. Reverse Engineer Anything (REA) is not likely to work with Claude.” A thread in the project’s own chat server read, “Claude is refusing to decompile anything,” and plenty of people chimed in that they’d hit the same wall.

On the OpenAI side it went further. An X user who goes by log4j said OpenAI banned their account for breaking its rules on “cyber abuse” after using Codex to take apart Samsung TVs. Others told 404 Media they had been warned or had their accounts restricted after asking Codex to look at hardware, or even at their own code. A separate site that made it easier to pick apart games and software, Decomp.wtf, shut down around the same time, though nobody has tied that directly to the SI blocks.

OpenAI, to its credit, didn’t dig in. Tarun Gogineni, an OpenAI engineer who posts as roon, answered a complaint from veteran programmer Eric S. Raymond with about as clear an admission as you’ll get from a big tech company.

“These blocks are mostly accidental and a result of difficult decision boundaries; there is no prohibition against reverse engineering with codex. it’s getting fixed,” he wrote. In plain English, the filter that’s supposed to catch criminals breaking into other people’s systems can’t always tell them apart from a guy fixing his own TV.

Claude’s brakes were installed on purpose

Anthropic’s story is different, because a lot of what users are hitting was built in deliberately. When the company launched its top model, Opus 5.5, on Sept. 22, it said “most cybersecurity tasks will be re-routed” to an older model. By Oct. 6 it was saying its public models “have conservative cyber safeguards that block most cyber work,” as Mavday’s review of the record lays out. Claude Code’s own help pages say the safety checks “most often flag cybersecurity and biology content,” and that they read the whole conversation, so one flagged request can poison every message after it. The fix Anthropic suggests is starting a fresh session.

Mavday found no announcement singling out Reverse Engineer Anything, and its count of complaints on Anthropic’s public bug tracker showed no sudden spike in late October. The blocks are real. They just look like a wide net that was already there, catching more people as the trend took off.

There is one rule Anthropic did add, and it’s about video games. Around Oct. 30, users spotted two new sentences in Claude Code’s instructions: “Do build original games, games in a genre, and games from the user’s own designs and assets. Don’t build a recreation of a specific existing game, even if it is renamed or called original.” Anthropic engineer Addy Osmani confirmed it, saying the company added the rule “because using Claude to infringe IP is against our usage policy,” and invited anyone whose own project got turned down to send it to him.

That rule overshot, too. Anthropic later told users it “was also kicking in when you referenced” a game “just to describe the mechanics you wanted,” and loosened it in a Claude Code update so you can name a game for inspiration and still get an original one, The Verge reported.

Owning your stuff is the whole fight

Here’s where a regular person should plant a flag. Copying somebody else’s game and selling it is theft, and SI companies are right not to help with that. Opening up a TV you paid for, so it stops tracking you and starts working again, is not. It’s the right-to-repair argument farmers have been making about tractors for years, and it’s the same argument here.

Anthropic’s own written rules seem to agree. Its updated usage policy, which takes effect Thursday, bans hunting for weaknesses in a system “without authorization of the system owner.” It also says the rules don’t prohibit “security research, testing, or tool development conducted on systems you own or operate.” If it’s your TV, you are the system owner. The problem is that a safety filter making split-second calls doesn’t know whose TV it is.

A safeguard notice that flagged Claude users have quoted says Anthropic is “improving these safeguards to reduce the amount of incorrectly flagged messages.” OpenAI says its fix is coming. Until both actually land, the hobbyists are voting with their feet. People doing this work told 404 Media they are moving to open SI models that run on their own computers, where nobody can close their account. One said he’s buying his own hardware because he doesn’t want to build on “an ecosystem that will eventually lock me out.” He added that the free models he’s turning to come mostly from China.

That should worry the big SI companies more than any headline. If the most hands-on, curious users in America decide the only way to fix their own gadgets is to leave American tools behind, that’s a loss nobody in Silicon Valley should want. The companies say they want SI that helps ordinary people do more. Letting a man fix his own TV is a good place to prove it.