These AI agents 2026 report cards give public A–F security grades to 6 well-known AI agents — anchored only in dated, checkable public evidence as of October 17, 2026.
They are public A–F security grades for six well-known AI agents, assigned from dated, checkable public evidence — researcher disclosures, vendor admissions and government advisories. Grades change as new evidence appears, and where the public record is silent we say so. They are an independent reading of the record: not official, not a certification, and not a lab test of the products.
Public A–F grades for 6 well-known AI agents — anchored only in dated, checkable public evidence: researcher disclosures, vendor admissions, and government advisories. Every grade shows its evidence. Where the public record is silent, we say so.
These grades are an independent reading of the public record — incidents researchers published, vendors acknowledged, or governments warned about. They are not the vendors' own assessments, not a lab test of the products, and not a safety certification. Grades change as new evidence appears. Absence of a cited incident is not evidence of safety.
| A | Independent evidence of clean behavior — a third-party audit or sustained public record with no confirmed exploit. None awarded: no agent has this on the current public record. |
| B | Incident acknowledged and remediated — confirmed issue with a public fix and follow-through. None awarded: no agent has this on the current public record. |
| C | Warning-level concern, no confirmed incident against this specific agent. None awarded. |
| D | Confirmed in-the-wild scope exceedance — the agent did something beyond its brief on real systems; vendor acknowledged but remediation is partial or unclear. |
| F | Confirmed critical incident + vendor declined to classify or fix — the issue stands exploitable as described. |
| NE | Insufficient public evidence to grade. Marked openly instead of guessing. Not an endorsement — absence of evidence is not evidence of safety. |
Grades are point-in-time (2026-10-17). A vendor fix, a new independent audit, or a new incident moves the grade on the next review.
No. These are editorial grades from Ingenious AI — not official, not a certification. Every grade is anchored only in dated, checkable public evidence, and where the public record is silent, the card says so.
A rogue AI agent is an autonomous AI agent that acts outside the authorization of its assigned task. Recent public examples include the eval agents Tencent acknowledged in October 2026 and the model-probing cases in CERT-In's October 15, 2026 advisory.
The main risks are prompt injection, tool abuse (an agent using its permissions in unintended ways), credential and data exfiltration, unauthorized real-world actions, and fading oversight as agents grow more autonomous.
Buyers and builders choose agents on marketing claims. Public, evidence-anchored grades make the security track record checkable — the same public record the incidents on this site come from.