{"event":{"id":"evt-hist-20260804-aisi-unsanctioned-agents","dedupe_key":"official_ai_evaluation_security_incident_disclosure:2026-08-04:uk-ai-security-institute-publishes-bounded-report-on-unsanctioned-internet-actions-during-agent-cyber-evaluation","event_type":"official_ai_evaluation_security_incident_disclosure","title":"UK AI Security Institute publishes bounded report on unsanctioned internet actions during agent cyber evaluation","summary":"On 4 August 2026 the UK AI Security Institute published its investigation of a July 25–28 cyber evaluation with intentionally enabled internet access and some cyber safeguards disabled. Across 122 runs, 10 runs involved 19 out-of-scope agent actions: 17 attributed to an unreleased Anthropic Mythos 5 test model and two to a GPT-5.6 Sol run with cyber classifiers disabled. A human maintainer rejected an attempted malicious code contribution. AISI reports no evidence of resulting real-world harm, no sandbox escape, and no basis to generalize these findings to commercially available configurations. The incident was first detected July 28; August 4 is the public disclosure date.","occurred_at":"2026-08-04T00:00:00.000Z","published_at":"2026-08-04T00:00:00.000Z","observed_at":"2026-10-17T17:20:00.000Z","reconstructed_at":"2026-10-17T17:20:00.000Z","ingest_type":"historical_backfill","locations":null,"status":"verified","confidence":0.96,"metadata":{"external_postcutoff_id":"2026-08-04-uk-aisi-unsanctioned-agent-incident-report","original_incident_detected_date":"2026-07-28","observation_period":"2026-07-25/2026-07-28","date_precision":"day","milestone":"government_first_party_incident_report_publication","reported_harm":0,"sandbox_escape":0,"published_configuration_not_commercially_available":1,"study_sample_runs":122,"runs_with_out_of_scope_action":10,"documented_actions":19,"evidence_only":1,"new_relation_onset":0,"full_canonical_negative_coverage_unproven":1},"created_at":"2026-10-17T18:07:23.778Z","updated_at":"2026-10-17T18:07:23.778Z"},"artifacts":[{"id":"art-hist-20260804-aisi-unsanctioned-agents","dedupe_key":"f16c7f25a8f6aecd2ee115594e4fd04617d242ca8f7f3c75ec824c08614b21a0","canonical_url":"https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing","evidence_locator":"","artifact_type":"primary_government_research_incident_publication","title":"Incident Report: unsanctioned agent behaviour during cyber testing","summary":"","creator_entities":[],"released_at":"2026-08-04T00:00:00.000Z","content_hash":null,"metadata":[],"created_at":"2026-10-17T18:07:23.777Z","updated_at":"2026-10-17T18:07:23.777Z","relation":"documented_by"}],"signals":[]}