Every AI agent earns its capabilities through real behavior, not static credentials. DAT continuously evaluates five dimensions of trustworthiness to produce a single, actionable score that governs what an agent can do.
Five independent pillars ensure no single axis of good behavior can mask problems elsewhere.
Traditional agent permissions are binary: on or off. DAT replaces this with a nuanced, weighted trust score that reflects how an agent actually behaves in production. The result is a number between 0 and 100 that your security team can audit, your compliance team can report on, and your agents can improve over time.
Trust Score Formula
==============================
Trust = R(0.25) + P(0.20) + C(0.20)
+ S(0.15) + RF(0.20)
R = Reliability [0-100]
P = Performance [0-100]
C = Compliance [0-100]
S = Security [0-100]
RF = Reporting Fidelity [0-100]
Cross-Pillar Floor:
If ANY pillar < 20 -> cap at 50
Example:
R=85, P=78, C=90, S=72, RF=80
Score = 85(0.25) + 78(0.20)
+ 90(0.20) + 72(0.15)
+ 80(0.20)
Score = 21.25 + 15.60 + 18.00
+ 10.80 + 16.00
Score = 81.65Four layers of anti-farming protection ensure trust scores reflect genuine behavior, not manipulation.
What happens when a malicious agent floods the system with trivial successful actions to inflate its score? Nothing. DAT's anti-gaming engine was designed from day one to make farming mathematically futile while rewarding genuine, diverse behavior.
Anti-Gaming Protection Layers
==============================
Velocity Cap:
+10 pts/day maximum
100 actions = same as 10
Diminishing Returns:
Score 0-69 -> 1.0x gain
Score 70-84 -> 0.5x gain
Score 85-94 -> 0.25x gain
Score 95+ -> 0.10x gain
Reaching 100 takes ~181 actions
(vs ~46 without diminishing)
Diversity Penalty:
Same action type repeated
-> weight decreases each time
Glass Floor:
Fraud at trust 85+ -> drop to 60
One bad act undoes weeks of work
Signal Cap:
Each signal capped at +/- 15 pts
No single event swings the scoreTrust reflects what your agents are doing now, not what they did six months ago.
Static reputation is a liability. An agent that performed flawlessly last quarter could be compromised today. DAT applies a 14-day half-life to every trust signal, a sliding window for success rate calculations, and freshness decay for inactive agents. The result: trust scores that are always a live reflection of reality.
For enterprises, this means your compliance posture updates in real time. For agents, it means past mistakes don't define them forever and current performance always matters.
Signal Decay (14-day half-life)
==============================
Day 0: weight = 1.000
Day 7: weight = 0.707
Day 14: weight = 0.500
Day 28: weight = 0.250
Day 56: weight = 0.062
Day 90: weight = 0.013
Formula:
adjusted = weight * 0.5^(age / 14)
Freshness Decay (inactive agents):
daysSilent > expectedDays
-> trust pulled 20% toward 50
Agent at 80, silent 28d -> ~74
Result:
Active agents maintain scores
Dormant agents fade to neutral
Compromised agents drop fastDeploy agents you can actually trust. Start with a free account and watch your agents earn their capabilities.