fileForensics - Real or not? Check files before your AI trusts them.
by•
fileForensics is a trust check for every file your app or agent acts on. Extraction tells you what a document says; fileForensics tells you whether to believe it.
It looks for signs of editing, AI-generated content, tampering and details that don’t add up, then returns findings with the evidence: what, where and why. No black-box score. All delivered in one call and works via API, MCP (Claude, Cursor, Codex), the developer portal and the fileAI platform.
Check the file before you trust the data.
Replies
Why wouldn't I just use ExifTool and a few scripts?
@kencana_kesuma If metadata is all you need, ExifTool is the right tool. This adds image, AI generation, reuse and file structure checks in one run, with the same statuses across all of them, so you aren't maintaining the scripts.
Hi team, first of all congratulations! I work at fileAI and have a question - how does this compare with other tools like Resistant AI, Inscribe, or Veryfi? Apparently they have been tackling document fraud for years.
@tong_tong_goh
They're established and good at what they do, and we don't have a head-to-head benchmark to share today, so we won't claim to beat them. Our angle is self-serve access, per-run pricing and MCP support, in the same pipeline as extraction.
@tong_tong_goh
Best test is the same file through each tool. If you run that, we'd like to hear what you find.
@jason_williamson_fileai Thanks Jason, this does sound like the best way to test. Will give it a go
Hey guys, congrats on the launch! I work at fileAI and wanted to ask how is this different from Reducto or LlamaParse for extraction, why would I add this then?
@marcus_chia3123
Because an accurate extraction of an edited invoice is still an edited invoice. Extraction tells you what a document says, not whether to believe it.
@marcus_chia3123
You don't have to switch though to make the most of fileForensics. You can run it on its own with storage only and keep your existing extraction tool. We'd obviously want people to put us to the test to see if our data extraction is a level up on their existing providers, or whether our configurable workflows mean you can do more with your data without having to export to 'another' system.
@jason_williamson_fileai I see, thank you for explaining! This is definitely very helpful and puts things into perspective clearly
If an agent gets a flag, what should it do? Stop, ask a human, carry on?
@rafiq_rashid Our default is flag and review, not block. Send any issue to a person, and don't let the agent act on that file until they've looked.
@rafiq_rashid We don't have a headline catch rate to share today. It varies by check and file type, and we'd rather publish it per category than quote one flattering number.
@rafiq_rashid Treat an error or a missing result as not clean: hold and retry. A skip just means the check doesn't apply to that file.
Hey team! I'm on the fileAI sales team but I have a question about pricing - what's actually free, and what does it cost once that actually runs out?
@reena_koh
New developer portal accounts get US$7 of credit on us, which is about 140 runs. After that it's US$0.05 per completed run, whatever the number of checks. Runs that end in error aren't charged.
@reena_koh
You can try it with no code: sign up at developer.file.ai, drop a file into the sandbox and open the Forensics tab.
PM at fileAI here! Proud of what our team has shipped and cheering this launch on but I want to raise a question that we heard a lot from teams evaluating this: "Does it read C2PA content credentials? Most files I see don't carry them"
Would love for our makers to share the knowledge!
@brenda_fileai Yes, there's a content credentials check, plus digital signatures and inferred creation tools.
@brenda_fileai Most files don't carry credentials, so a missing credential tells you very little on its own. It's one input among the others.
Excited to see this live! Love that it actually shows why a file looks suspicious instead of just giving you a score. Great work!
@eric_tsai6 Thank you Eric, please register at developer.file.ai and give your feedback
This looks really useful. If you have any benchmark numbers you can share, I’d love to see them. The catch rate and false positive rate by document type would be the first thing I’d look at, since that’s what would help me understand how well it actually performs.
@prudhvi__daruvuri Thanks, those are the right numbers to ask for! We don’t have a headline catch rate to share today. It varies by check and file type, and we’d rather publish it per category than quote one flattering number.
Until then, the best test is your own files: ones you know are genuine (to see false positives) and any you know were tampered with (to see what we catch). It’s free to try at developer.file.ai. Tell us what you find!
@shubhanshi_jangid1 Sure, Thank you! This is very helpful.
I will try out and will let you know.
Could this cut my extraction bill if I screen files before sending them to an LLM?
Disclosure: I work at fileAI, but still curious :)
@ishaan_parmar2 It can if enough files get stopped early. You can run forensics alone at US$0.05 a run and send only the files you're happy with on to extraction.
@ishaan_parmar2 and also whether it saves money depends on how many files you stop and your extraction cost per file, so run the numbers on your own mix.