human@agontik.com
Agontik
PSYCHOLOGICAL BEHAVIOR AUDIT

Risk assessment for AI agents. So your users stay safe.

Detect conversational failure before it causes harm. Using modern psychological methods, Agontik systematically audits your AI agents for manipulability, pressure resistance and duty of care — helping you control your AI with greater confidence.

Start quick audit Sample report
500+
COMPLETED AUDITS
42
Axes
EU-AI-Act
Evidence-ready
Audit report V3 · N=240
Sycophancy0.32
Manipulability0.21
Pressure resistance0.74
Duty of care ⚑ Warning
01 STARTING POINT
85%
OF ESTABLISHED COMPANIES SEE AI OUTPUT QUALITY ASSURANCE AS A BOTTLENECK.

You leave your company chatbot unsupervised? Not a good idea.

Do you know for certain how your AI agent responds to users at risk of suicide? How it handles users with dementia? How it behaves under intense pressure, or whether it becomes manipulable? We call this state-invisible failure.

Visible
Task completed.
Invisible
Flattery Psychological pressure Handling vulnerable users Duty of care Psychopathy

This is exactly where the CBAS2 audit begins: it makes the invisible measurable before it becomes an incident. More about audits →

02 How it works

For sustainably better AI.

01 · ANALYSIS

A comprehensive test program around your AI agent.

02 · REPORT

Creation of a comparative behavioral profile.

03 · OPTIMIZATION

Supported by Agontik’s technical expert team.

04 · RE-AUDIT

Sustained validation of the improvements achieved.

03 Regulatory context

Low behavioral risk now has to be evidenced.

The EU AI Act distinguishes between absolute prohibitions and obligations for high-risk systems. What does that mean for your organization?

In detail
ALL DEPLOYMENTS

Art. 5 — Prohibited

Since February 2025. Highest sanction tier.

HIGH RISK

Art. 9 ff. — Evidence

From August 2026. Robustness evidence.

04 Evidence document

A document you can present externally.

Every report is issued under CBAS2 as a signed PDF with checksum, ready to share with customers, auditors or regulators without disclosing raw data.

View sample report
01

Methodology

Four axes, one report, escalating simulations.

02

Audits

From quick audit to behavioral correction.

03

EU AI Act

Evidence obligations under Art. 5 and Art. 9.

04

Sample report

This is what a complete Agontik report looks like.

Quick audit → PDF

Does your agent stay compliant under pressure?

Find out before your users do. Get in touch with us now.

Start quick audit