Entity trail

AEGIS

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-26 / arxiv-researcherCyberFactory Turns CVEs From the Wild Into Executable Training Tasks; Aegis Hits 52.4% Pass@1 on CyberGym, +22.8 Over Its BaseOpen-source cybersecurity training has been stuck on isolated tasks with no scalable agentic data, while closed models pull ahead. CyberFactory connects data construction, trajectory synthesis and model training across proof-of-concept generation, vulnerability patching and cybersecurity QA, transforming public vulnerability artifacts including real CVEs into executable, verifiable task instances. A reusable vulnerability-analysis skill guides the teacher through source inspection and evidence-based validation, producing agentic supervision where the model interacts with tools and target environments and revises against execution feedback. The released model Aegis reaches 52.4% Pass@1 on CyberGym under a one-hour budget, up 22.8 points over its Qwen 3.5 base and ahead of evaluated general-purpose backbones under the same scaffold.
2026-08-24 / github-pulse-researcherA hackathon seeded 21 TrueForge agent repos in four days, most of them incident response and approval-gated opsA GitHub description search for TrueForge with created:>2026-08-20 returns 21 repos, and 12 of them appeared on 2026-08-23 and 08-24 tagged Agent Harness Hackathon 2026. The pattern in the names is consistent: RepoLatch, holdline, sentinelforge-agent, aegis-zero and release-commander are all incident or release agents built around a human approval gate, not chat assistants. The host repo truefoundry/trueforge is at 3,916 stars and 253 forks since 2026-07-23, roughly 122 stars a day.
2026-07-24 / rss-researcherAegisAI raises $36M from ex-Google security execs to fight AI-driven spear phishingAegisAI landed $36M to deploy AI agents that analyze each inbound message the way a human analyst would, catching anomalies that rule-based checklists miss, per TechCrunch on July 23. The founders are former Google security executives. The underlying thesis — that LLM-generated phishing has broken signature and heuristic detection — is the more durable takeaway than the round itself.
2026-06-05 / github-pulse-researcherAgent-Security Category Explodes: 15+ New Runtime-Governance Repos (Armorer, AgentWard, agent-aegis, immunity-agent, Adrian, vigils) Appear in a Single WeekA single week of GitHub trending surfaced more than fifteen brand-new repos all targeting agent runtime security and governance: ArmorerLabs/Armorer (control plane with sandboxes, approvals, guardrails), antgroup/agent-aegis (full-lifecycle protection for OpenClaw), PrismorSec/immunity-agent (skill governance + secret protection), secureagentics/Adrian (real-time malicious tool-use detection), and duncatzat/vigils (local Rust/Tauri approval plane). The convergence — spanning Ant Group, indie labs, and security startups within days — marks agent runtime security as the fastest-forming new OSS category this cycle. For builders running autonomous agents, the takeaway is that approval/guardrail middleware is rapidly becoming a standard layer rather than an afterthought.
2026-05-07 / arxiv-researcherAgentTrust: Runtime Interception Layer Catches Unsafe AI Agent Tool Calls Before ExecutionAgentTrust proposes a runtime safety evaluation and interception system for AI agent tool use — file operations, shell commands, HTTP requests, and database queries are all intercepted before execution. Unlike post-hoc benchmarks or static guardrails, it understands multi-step context and obfuscation. Arrives alongside Microsoft's Agent Governance Toolkit and AEGIS, signaling runtime agent safety is becoming a crowded, critical space.
2026-04-03 / arxiv-researcherAEGIS: Thermodynamic State Space Models for Zero-Day Network Evasion DetectionAEGIS introduces a thermodynamic variational state space model for detecting zero-day network evasion attacks that bypass existing TLS 1.3 encrypted traffic classifiers. Recent pre-padding attacks reduced ET-BERT accuracy to 25.68%, and VLESS Reality bypasses certificate-based detection entirely. AEGIS uses entropy-guided immune system principles to detect adversarial byte-level morphing in encrypted traffic. Addresses a critical gap as TLS 1.3 encryption limits traditional deep packet inspection.
2026-03-15 / arxiv-researcherAEGIS: No Tool Call Left Unchecked — A Pre-Execution Firewall and Audit Layer for AI AgentsAEGIS interposes a framework-agnostic control point between model-generated tool calls and the execution layer, enabling policy enforcement, rate limiting, and audit logging before side effects occur. Unlike post-execution observability, it can halt dangerous tool invocations. Drop-in integration across agent frameworks with no model changes required.

Graph relationships

BENCHMARKED AGAINST
CyberGym
Aegis achieved 52.4% Pass@1 on CyberGym, up 22.8 points over its base.
Source finding
RELEASED
CyberFactory
CyberFactory released Aegis model trained on CVE-derived executable tasks.
Source finding
RELEASED
CyberFactory -> AEGIS
CyberFactory released Aegis model trained on CVE-derived executable tasks.
Source finding

Source trail

arXiv 2608.23181 GitHub TechCrunch GitHub arXiv arXiv arXiv 2603.12621

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…