A hacking group called “The Com” released personal information of hundreds of federal agents, including 680 DHS employees, 190 Justice Department officials, and 170 FBI employees[1]. The hack revealed names, office locations, and home addresses of agents.
The incident occurs amid heightened tensions, with ICE claiming a dramatic rise in assaults against officers, though these statistics are disputed. Colorado Public Radio found only a 25% increase in assaults rather than the administration’s claimed “1000% increase”[2].
The Department of Homeland Security under Secretary Kristi Noem promised prosecutions against those who dox ICE agents in July 2025, but three months later no charges have been filed[2:1]. This contrasts with California, where prosecutors have charged three activists for allegedly following and broadcasting an ICE agent’s home address[2:2].
The hack comes as the Trump administration has pushed tech companies to remove ICE-tracking apps, with both Meta and Apple recently complying with requests to take down such tools[2:3].
TL;DR: positive experience using AI code gen. Downvote button is that way (clients may vary) if that’s all someone cares about before voting ---->
I’ve definitely reviewed code and had to tell people not to use AI because A) depending on what proprietary files or information they sent to the AI it was literally illegal and B) it was garbage code with random bits of oddness that the developer couldn’t explain.
That said, there’s a bit of a skill issue here.
I’ve been experimenting with generated code at work (because it’s literally my job to enable my team in this way) and without knowing the code base at all but understanding a series of endpoints that needed to be called, I spent about 6 hours researching and documenting all the service calls and schemas, explaining the command line options I needed and how they would function. (Existing standards and docs are inherited garbage—I’m sure you know how it is.)
Anyway I put in a bunch of work explaining what needed to be done and how to do it and just let Cline and Claude cook. Ten minutes later I had working code and a battery of unit tests to validate it was correct. I caught a couple of logic errors reviewing the generated code and rejected it with an explanation of what it had done wrong. It tried a second time and nailed it.
It retrieves an auth token, parses and validates minimum data in a spreadsheet, checks to see if the users are okay to overwrite their login information, then submits updates to the account, waits/polls another system that has to ingest that update through Kafka, validates the ingestion was successful, and finally creates/overwrites a login for the user. Various commands line options execute subsets of that. It logs every API call inbound and out, logs the success of each line item or where it failed, and also lets you pass in a filter file to only process some of the lines of the spreadsheet.
It also wrote about 60 unit tests which did identify a bug or two and it went back and fixed them on its own. In an ecosystem I’m only beginning to learn and in a language I can’t write. I am a well-seasoned SSE/Tech Lead, though, so it’s not like I don’t know what I’m doing.
And I can say there were some edge cases in the data that tested almost every failure mode. One account couldn’t be overwritten, a couple were missing required fields, the Kafka consumer fucked up and we had to help that team troubleshoot. The script was rock solid.
In the right hands, using the right tools, AI can be a good coding partner. It’s also the best / most expensive coding model (that I’m aware of) and the AI ate up about $15 in API costs. Which us a bargain compared to what they would’ve paid me to write it.