2026-11-22 16:54 UTC

Together AI says exposed development credential enabled prompt-sample downloads

state: seedheat: highuncertainty: mediumrelevance: hightogether-ai prompt-leakage provider-securityTogether AIVipul Ved Prakash

What is this?

Together AI, an inference provider, disclosed on November 21, 2026, that an unauthorized party used a credential exposed in a public code repository to download stored prompt samples from a private Hugging Face repository. The samples came from a subset of serverless requests to MiniMax M3, GLM-5.2, and Kimi K2.6, primarily on four dates in June and July 2026. Together AI says an AI coding agent committed the credential while the company was developing its public xorl repository; automated scanning detected the exposure and the commit was reverted, but the credential was used during the exposure window. The company says it is still reviewing the prompts to determine what sensitive information was exposed and which customers need direct notification. It reports changes to credential handling, pull-request scanning, default retention settings for new accounts, and AI-agent workflows. The details and scope come principally from Together AI’s own notice; the supplied sources do not independently establish the extent of customer impact.

Why it matters

This is a material privacy and security incident for AI inference infrastructure: prompts submitted to a provider were stored for testing and later downloaded by an unauthorized party after an exposed development credential. It illustrates how AI coding agents can introduce risk through access to credentials, even when routine scanning catches a leak. Together AI’s review is unfinished, so the sensitivity and customer impact remain unclear. Its reported move to make zero data retention the default for new accounts may improve protection for future users, but does not by itself resolve what happened to the already downloaded samples.

Current assessment

New: Together AI’s November 21, 2026 first-party disclosure says an unauthorized party downloaded stored prompt samples associated with a subset of serverless inference requests from June and July 2026. The samples were collected for routine software testing and stored in a private Hugging Face repository; the company says an AI coding agent committed a credential granting access to that repository while developing its public xorl GitHub repository. Automated scanning detected the exposed credential and the commit was reverted, but the company says the unauthorized party obtained and used the credential during the exposure window. The affected samples came from requests to MiniMax M3, GLM-5.2, and Kimi K2.6, primarily on June 24, July 4, July 8, and July 14. The notice says the party searched the samples for credentials to other services and tested whether they were valid; five tested credentials were canaries associated with a Dragos security experiment. Together AI reports working with an external forensic firm and law enforcement, says it has found no evidence of additional prompt leakage beyond this incident, and says review of affected prompt contents is ongoing. It has not yet established in the notice whether sensitive personal information or other customer data was exposed, or the full set of affected customers. Reported changes include scoped or short-lived credentials and keyless access where supported, credential checks on all pull requests, zero data retention as the default for new accounts, and revised development processes for AI coding agents. The company advises customers handling sensitive prompts to enable zero data retention. Implications: prompts sent through hosted inference may be retained for operational or testing purposes unless retention settings prevent it; access-control and agent-workflow safeguards matter alongside model security. Track the company’s eventual impact findings and customer notifications. The source evidence is mainly the company’s own disclosure; secondary search results largely repeat it and do not independently confirm scope.

Measured heat

now 0 pts/hpeak 4 pts/hcomments 0/hpeers p30momentum: steady2 platformsage 51h
points/hour across evidence · reading as of 2026-11-23 03:23:31.420709+11:00 · deterministic, not a model opinion

How the heat travelled

11-20 13:00⭐ origin echo-reconstructedThis is the original primary source, not an intermediary: Together AI’s disclosure says an AI coding agent committed a credential for a priv
Vipul Ved Prakash on blog (echo) · attributed from hn.story.50595816
—
11-22 03:07first on hacker news · published · +38.1hTogether.ai: Prompt Leakage for Requests
Fe2O3
—
11-22 03:07amplified on hacker news 👑hn.story.50595816
Fe2O3
peak 19 · 7 comments · 100% of case engagement
11-22 07:20our radar first saw it · +42.4hdiscovery anchor: hn.story.50595816—
pace: p72 vs 1999 stories at the 48h mark (now 51h old) — ahead of postman-agent-plugin-install-consent-claim (1.0x), behind amazon-engineering-role-ai-guidelines (1.0x)

Evidence (2) — ⭐ canonical anchor

sourceobjectauthorscorecomments
🟧 hnTogether.ai: Prompt Leakage for Requests
Retrieved article excerpt

Open article · Retrieved 2026-11-22T08:22:15.030960+00:00

We are notifying customers about a security incident involving a subset of prompts submitted to Together AI’s serverless inference service in June and July 2026. An unauthorized party downloaded stored prompt samples after a credential for a private repository was exposed in a public code repository.

We apologize for this incident. We are reviewing the affected data to determine what information was exposed and which customers require direct notification.

## What happened

As part of routine software testing, we collected prompt samples authorized for that use and stored them in a private Hugging Face repository. The samples came from requests to the MiniMax M3, GLM-5.2, and Kimi K2.6 models, primarily on June 24, July 4, July 8, and July 14, 2026.

During development of our public GitHub repository, xorl, an AI coding agent committed a credential that provided access to the private Hugging Face repository. Our automated scanning detected the exposed credential, and we reverted the commit.

During the exposure window, an unauthorized party obtained the credential and used it to download the stored prompt samples. The party subsequently searched those samples for credentials to other services and tested whether those credentials were valid. Five of the credentials tested were canary credentials associated with a Dragos security experiment.

## Investigation and scope

We have worked with an external forensic investigation firm and law enforcement to investigate the incident.

Based on our review, we have not identified evidence of additional prompt leakage beyond this incident. Our review of the affected prompt contents is ongoing to determine whether sensitive personal information or other customer data was exposed. We will contact affected customers directly as needed with the findings relevant to them.

## Changes we have made

Following the incident, we strengthened controls around development workflows, credentials, and customer data. These changes include:

1. Using scoped or short-lived credentials, and keyless access where supported, to reduce the potential impact of credential exposure.
2. Checking all pull requests for credentials.
3. Making zero data retention (ZDR) the default setting for new accounts.
4. Updating our development processes governing the use of AI coding agents.

## What customers can do

Customers handling sensitive prompt data should enable ZDR in their account settings.

If you have questions about this incident or its potential impact on your organization, contact [email protected].

‍

**Vipul Ved Prakash**

CEO and Co-founder, Together AI
Fe2O3197
🟧 echo.blog ⭐This is the original primary source, not an intermediary: Together AI’s disclosure says an AI coding agent committed a credential for a privVipul Ved Prakash——

Interpretation history

Decision trace