Voice intake for global research teams
AI researchers and safety leads often work in mixed-language teams. Reporters can dictate their report in any of 17 languages. The server only ever sees the text.

AI companies face a specific version of the EU whistleblowing directive: global research teams, multilingual staff, and reputational stakes that are higher than most industries. We give you a channel that meets Directive 2019/1937, fits how AI teams actually work, and passes a real procurement due-diligence questionnaire.
5 min
From signup to a working reporting channel.
17
Locales for global research teams. Voice intake, no separate per-language site.
0
Audio recordings stored. Audio is transcribed in the browser, then deleted. We never see it.
E2E
End-to-end encrypted. The reporter holds the key. We cannot read the report contents.
Overview
The EU Whistleblower Directive 2019/1937 applies to any legal entity with 50+ employees that operates in the EU, regardless of industry. AI companies hit this threshold early — often at Series A — and once they sell to enterprise customers, the requirement is reinforced by procurement due-diligence questionnaires that ask for a documented internal reporting channel. The AI Act (Regulation (EU) 2024/1689) does not itself require a whistleblower channel, but it does require general-purpose AI providers to have internal grievance mechanisms for fundamental-rights issues. A single channel covers both obligations without operational overlap.
What applies to you
Most AI companies we work with are navigating two parallel compliance regimes at once. Here is what the law actually requires — and where our channel maps onto it.
Internal reporting channels for legal entities with 50+ employees. Seven-day acknowledgment, three-month feedback, no retaliation. Member-state law (e.g. HinSchG in Germany, Loi Sapin II in France) layers on local specifics.
General-purpose AI providers must enable fundamental-rights complaints. Our internal channel routes those to the right team without changing the data flow.
Enterprise customers routinely include a "describe your internal reporting channel" question in their vendor security questionnaires. Saying "email us" is not a passing answer in 2026.
Anti-retaliation duties apply to the AI company once a report is filed. Documented intake + case history is the employer's strongest defense in any subsequent dispute.
AI companies pursuing these certifications need a documented, auditable intake and case-management process. Our per-case event log with timestamps and acting users is the audit trail that supports these requirements — it isn't the whole audit, but it's the most-tested part.
Why Ashio for AI
The reason most AI companies pick us over a generic ticketing tool or a Slack-channel workaround.
AI researchers and safety leads often work in mixed-language teams. Reporters can dictate their report in any of 17 languages. The server only ever sees the text.
Every case gets a unique URL the reporter can use to follow up anonymously. No login, no account, no friction. Same model as how Linear and Notion share issues.
The reporter encrypts the report in the browser with a key only they hold. Your safety lead decrypts. We cannot read the contents.
Even below the directive's 50-employee threshold, a documented internal reporting channel helps you pass enterprise procurement questionnaires. Many of our customers under 50 staff deploy it pre-Series A for the SOC 2 audit, or because their first enterprise customer asked for it on the vendor security questionnaire.
Reports can be auto-assigned to your safety lead, your DPO, or your compliance officer based on the report category. No manual triage required.
The case list filters by category — model bias, data privacy, eval integrity, and others. Every report carries its category from submission, so your safety lead can pull only the relevant items.
Compliance checklist
A pre-flight list for AI companies about to sign a contract that requires a documented reporting channel.
Most procurement questionnaires ask for this as a hard requirement. A live URL with a unique access link per case is the standard answer.
The directive requires feedback to the reporter within seven days. We enforce this with a per-case deadline that is visible to every admin from day one.
The directive also requires feedback on how the report was handled within three months. Our case status supports this out of the box.
The directive prohibits retaliation against the reporter. Your hiring, compensation, and assignment records need to be defensible. The case log is the source of truth for that.
If you have staff in multiple EU countries, the local transposition may add extra requirements. Our locales system maps each country to its specific law.
FAQ
No, not directly. The AI Act (Regulation 2024/1689) requires general-purpose AI providers to enable fundamental-rights complaints under Article 86. A whistleblower channel is required by the separate Directive 2019/1937, which applies to legal entities with 50+ employees. Most AI companies we work with operate the single internal channel that satisfies both — the legal duties are complementary, not duplicative.
Not for EU Directive 2019/1937 compliance. But if you sell to enterprise customers, the SOC 2 / ISO 27001 questionnaires most large buyers use will ask "do you have an internal reporting channel?" Saying "no" is increasingly disqualifying. The cost of a 14-day free trial is much lower than losing a deal over that question.
Yes. The intake is a public URL — reporters can file from any browser, in any country, in any of 17 languages. The dashboard is a normal web app. There is nothing to install on the reporter's device or in your infrastructure.
Our role is the intake and the case log. The actual anti-retaliation protection (employment law, HR investigations) is your responsibility. What we do provide is a defensible record: every case has a full event log with timestamps and the acting user, so any subsequent retaliation claim can be cross-referenced against the case timeline.
Yes. You can set a default assignee per case category, so reports about model bias, data privacy, or eval integrity go straight to the right person. Multi-stage review (initial triage, then escalation) is also supported.
Yes. The EU Whistleblower Directive (Article 10(2)(b)) explicitly lists "voice-to-text facilities" as a compliant channel. Our voice intake transcribes in the reporters browser, then deletes the audio. The server only ever sees the resulting text transcript.
Within 7 days of receipt. EU Directive 2019/1937 (Article 7) requires acknowledgment within 7 days, and feedback on the outcome within 3 months. Our platform enforces both deadlines as visible timers per case.
Most SOC 2 audits look for three things: (1) a documented intake with an accessible URL, (2) an acknowledgment SLA with timestamps, and (3) a tamper-evident log of who handled the case. We have all three as part of the standard product.
For the private sector, the directive applies to legal entities with 50+ workers. For the public sector, it applies to all entities with 10+ municipal workers and all government bodies regardless of size.
Related reading
Start a 14-day free trial. The compliance is included, the audit log is included, the encryption is included.