You Gagged Our App for Trauma Survivors. Now Real People Are Suffering

We’re building a support tool for trauma and rape survivors, like people who often have literally no one else to turn to. Our app lets them write out their stories in their own words and receive detailed, legally-structured reports they can take to police, legal counsel, or therapists.

For many, this is the first time they’ve been able to tell their story without fear or shame. It’s a bridge to justice and healing for people who’ve been really screwed.

Until recently, this worked. We have our own safety filtering layer, but when users chose to disable them (consensually, for their own healing), the system would still meet them with compassion, structure, and clarity.

But now, something has changed with the new Gemini-Pro/-Preview-05-06. Even with every filter disabled, the model won’t talk to them anymore. It shuts down, refuses to engage, or produces hollow boilerplate that’s worse than nothing.

Victims are being re-silenced by the very tool that was built to give them voice.

We’re watching the damage unfold in real time. Survivors reaching out for help, hitting a wall of “I can’t help with that. bullshittery.

Therapists and support workers who had begun integrated this tool into their process are now up our ass, scrambling for answers. We are not a novelty use case.

We are not some throwaway edge case. This is a core, humanitarian use of this technology, and it’s being locked out.

We’ve tried to play by the rules. We’ve used content filters responsibly. We’ve documented use and intent. We’ve escalated internally and through all the right channels.

I have had several meetings with internal Google staff, we were in the middle of setting up a five figure monthly contract, and now the rug has been completely pulled on us.

So now we’re posting here.

If you’re at Google, OpenAI, Microshit, whatever, or working on these systems, please understand: this isn’t just about alignment or content policy. This is about human lives and critical support that’s vanishing without warning.

We’re happy to provide examples, redacted stories, anything that helps this get the attention it deserves.

Please help us fix this. If anyone knows anyone at Google, or any way to get this across, I would be so appreciative.

So far we’ve been approached by The Register, and other news media about this, and if we could get some answers that would be very appreciated

@Logan_Kilpatrick, anyone.

Please.

5 Likes

This is yet another sad example of the real-world fallout from Google’s recent silent redirect of the clearly dated “gemini-2.5-pro-preview-03-25” endpoint. Removing access to a previously working model without warning is already problematic (even if technically allowed by “preview” policy). But quietly, without warning, redirecting it to a new, unpredictable model whose behaviors developers can’t possibly anticipate causes real harm to those who rely on stable checkpoints to validate, build out, and support critical use cases exactly like yours.

Your story highlights precisely why we’ve been pushing Google to publicly acknowledge and address this issue. It’s not even about developer inconvenience or technical frustration… real people are being harmed by these unexpected changes.

If you’re comfortable doing so, please consider adding your voice to this thread calling on Google to restore the original “03-25” snapshot endpoint and/or clarify their policy around dated checkpoints going forward:

Thanks again for sharing this. It’s important that Google sees the human consequences of these decisions.

2 Likes

Thank you for your solidarity, and support.

Really appreciate it, hopefully we can get some change made.

Added my voice to the thread too.

2 Likes

To my knowledge, gemini-2.5-pro-preview-05-06 is an experimental model designed solely for feedback and testing purposes, not intended for production use.

1 Like