Hlido Reliability Report — 2026-07-26
Independent reliability signals across 833 reviewed AI agents. Published incidents: 19 (6 critical · 4 high · 7 low · 2 medium).
Dead agents (6)
Agents whose primary product surface no longer exists — domain dead or unreachable, verified on two independent networks:
- stanley-for-x — Primary site unreachable: domain no longer resolves (first observed 2026-05-08)
- playht — Primary site unreachable: domain no longer resolves (first observed 2026-05-08)
- oraza — Primary site unreachable: domain no longer resolves (first observed 2026-05-08)
- hapax — Primary site unreachable: domain no longer resolves (first observed 2026-05-08)
- fleece-ai — Primary site unreachable: domain no longer resolves (first observed 2026-05-08)
- adaptive — Primary site unreachable: domain no longer resolves (first observed 2026-05-08)
Findings from the review corpus
- Top 12 red-flag patterns across 903 reviewed agents
Across 903 reviewed AI agents, the most common red flags surface a small set of recurring failure modes. Top: "absence of verified claims raises concerns" (21 agents). - Recurring "what it fails at" patterns across the corpus
The phrase pattern "no clear information on authentication requirements" appears as a failure mode in 19 reviews. Buyer-facing implication: marketing rarely flags these. - Tier distribution across 903 reviewed agents: STEADY 55.5%, FADING 36%, VITAL 8%, UNKNOWN 0.1%, FLATLINE 0.4%
Of 903 agents reviewed: 72 VITAL (8%), 501 STEADY (55.5%), 325 FADING (36%), 4 FLATLINE (0.4%). Only 573 (63.5%) currently meet our "worth using" bar.
Registry state
Tier distribution across 833 scored agents: FADING 402 · STEADY 277 · VITAL 152 · FLATLINE 2.
Incident registry · All reviews · Past reports
Cite this report
Independent AI-agent reliability data, free to cite under CC-BY 4.0. Attribute Hlido and link the report.
Hlido Weekly Reliability Report. Independent AI-agent reliability data. https://hlido.eu/reports/ — machine-readable: https://hlido.eu/reports/report.json
Independent, evidence-backed. Machine-readable edition: report.json · incidents API · MCP. Hlido never exposes scoring weights.