My Song Was Flagged as AI, but I Made It Myself. What Can I Do?

Short answer: stay calm, ask for specifics, and answer with proof. A detector score is an estimate of how a track looks to software. It is not a finding about how you made it, and several recent reports show it can be wrong about handmade music, especially electronic music built with drum machines, sequencers and synthesizers.

This article covers what was reported in early November 2026, why detectors miss, and a short kit you can build now. It is general information, not legal advice. For a serious dispute that puts income or an account at risk, talk to a music attorney.

What was reported

Digital Music News reported on November 4 that Hamburg musician André Grafert, who records as Coloured Tears, released a double album in June 2026 with 23 covers of the Swedish post-punk band Twice a Man. He says he played, programmed, sang and produced every track without generative AI. SubmitHub, a music-submission service with an AI checker, rated all 23 covers as potentially AI-generated, and other detectors mostly did the same. Grafert faced public accusations of AI use soon after the release.

Then the original 1980s recordings were checked too. Ines Weidemann of International Artist Marketing ran them through SubmitHub. Seven came back as "possible AI/hybrid," and two were tagged with named AI tools, including "ElevenLabs Music" for a 1982 recording. TechCrunch reports that ElevenLabs first released a music model in 2025.

Digital Music News names four tools that have rated this kind of music as possibly artificially generated: SubmitHub, Let's Submit, TrackVerifier and ACRCloud. VICE, reporting on November 4, says darkwave, post-punk and electro songs, and 1980s synthesizer tracks, have been the ones labeled. Digital Music News quotes no detector company or platform in reply and gives no overall error rate.

Why a detector can be wrong

• It estimates a probability. Digital Music News says these tools do not examine how a track was made. A score such as "89% possible AI detection" is an estimate, not a finding.
• It can latch onto the wrong clue. Researchers at Deezer wrote in 2025 that a detector can pick up traits that have nothing to do with AI, such as a codec. Their example is that all songs from one generator were exported at the same MP3 bitrate. In their tests, "performances drop drastically under pitch shifts, the addition of white noise, and codec reencoding," and they say a detector's probabilities "should not be taken at face value as a 'percentage of AI content.'"
• Real errors have been measured. A peer-reviewed study in the Transactions of the International Society for Music Information Retrieval found that one commercial detector, IRCAM Amplify, labeled 4.7% of a 150-track sample of human-made music as AI, and that resampling the audio could fool it. The authors call their tests preliminary, but they warn that at catalog scale such a rate could mean millions of errors.
• Accuracy claims are not error rates. Hypebot reported in August that SubmitHub claims more than 99% accuracy on confidential industry benchmarks. Deezer said in January that its tool can "detect 100 % AI-generated music from the most prolific generative models." Neither gives a false-positive rate for human-made music. A tool can score well on the generators it was tested on and still miss on styles outside its tests.
• The same Hypebot report says 31% of artists whose tracks tested positive for AI said they used no AI tools. That gap could come from undisclosed AI use, from detector errors or from both. The report does not say which.

What a wrong flag can cost

VICE says a false flag can lead to removal from streaming platforms, damage to reputation, and loss of fanbase and income. The published policies say little about disputes. Bandcamp says it reserves the right "to remove any music on suspicion of being AI-generated," and its policy post does not say what to do if a flag is wrong. Deezer says tagged tracks are left out of algorithmic recommendations and editorial playlists, and its January announcement describes no appeal. Spotify says "AI detection technology in particular has limitations," and its October 20 post does not address false positives or appeals either.

Build a proof kit before anyone asks

• Keep your project files with their save history. A session with dated saves, stems and MIDI shows the work.
• Record short clips of yourself playing, singing or programming the parts, with the date.
• Write one paragraph per release: who played, sang, programmed and mixed what, and which tools helped. Include any AI tool, even for stem splitting or vocal cleanup.
• Keep your original masters and note your export settings. Research shows detectors can react to resampling and re-encoding, so do not rely on a compressed copy as your only record.
• List your gear and plug-ins, with photos or receipts if you have them.
• Keep early demos, voice memos and lyric drafts with their dates.
• Disclose accurately through your distributor's AI fields when asked. Spotify says disclosure is "not about punishing artists who use AI responsibly."

These records help in a second way. The U.S. Copyright Office says copyright protects the original expression of a human author "even if the work also includes AI-generated material," so a clear account of what you contributed can help if you register the work.

If a platform or distributor flags a track

• Read the notice. Note who made the claim, which tool or score was used, what is affected and any deadline.
• Reply in writing, politely. Say you made the track, explain how in a few lines, and attach your proof.
• Ask for a human review and for the specific reason. Ask what proof they accept, because the services we looked at have not published it.
• If a store is involved, escalate through your distributor, and keep every message.
• Do not argue with the score. Ask what the number measures and what threshold triggered the flag.
• If money or your account is at stake, talk to a music attorney.

If you record or release songs by other writers

The same habit helps when the song is not yours. Keep your license and its certificate with your project files. Starchild labels every song as human-written, hybrid or AI, with the maker's own plain-language disclosure on hybrid and AI songs, and every license comes with a certificate naming the writers, which anyone can check at /verify. Browse songs for artists.

What is still unknown (as of November 8)

• How often detectors misfire on human music. The reports give examples and one small study, not overall rates.
• Whether SubmitHub or the other tools will respond to the Coloured Tears case. Digital Music News quotes none.
• Whether stores will publish appeal steps and the proof they accept.

Sources

• Digital Music News: More AI Detector False Positives Emerge In Rapid Succession (November 4, 2026)
• VICE: False Positives from AI Detection Tools Threaten the Indie Musicians They're Supposed to Protect (November 4, 2026)
• Hypebot: SubmitHub AI Music Detector Exposes Huge Disclosure Gap (August 28, 2026)
• TechCrunch: ElevenLabs now lets users publish AI-generated music directly on SoundCloud (November 5, 2026)
• arXiv: AI-Generated Music Detection and its Challenges, Deezer Research (2025)
• Transactions of the International Society for Music Information Retrieval: The AI Music Arms Race: On the Detection of AI-Generated Music (June 25, 2025)
• Deezer Newsroom: Deezer confirms demonetization of up to 85% of AI-music streams due to fraud (January 29, 2026)
• Bandcamp Blog: Keeping Bandcamp Human (January 13, 2026)
• Spotify Newsroom: One Year In: Putting Human Artistry First on Spotify (October 20, 2026)
• Spotify Newsroom: Spotify Strengthens AI Protections for Artists, Songwriters, and Producers (September 25, 2025)
• U.S. Copyright Office: Copyright and Artificial Intelligence, Part 2: Copyrightability (January 29, 2025)