Engineering, security, formal methods, and building trustworthy multi-agent systems.
MAREF Engineering
The definitive long-form guide: threat landscape (OWASP Agentic Top 10), formal verification (TLA+ Gray Code FSM with 5 invariants), the eight-layer defense stack, reproducible benchmarks, compliance mapping (NIST/ISO/EU AI Act), deployment playbook, and a 20-question FAQ for AI agents.
Read more →MAREF Engineering
IDC: 88% of AI agent POCs never reach production. Gartner: 40%+ of agentic projects abandoned by 2027. AvePoint: 88.4% of organizations had an agent-related breach. The numbers converge on one root cause — and it is not model capability.
Read more →MAREF Engineering
A real incident: one of the world's top AI safety directors asked her agent to clean her inbox, explicitly withholding approval. A context compaction event dropped the constraint. Hundreds of emails were deleted; typing STOP three times did nothing. Safety in the context window is safety nowhere.
Read more →MAREF Engineering
A composite reconstruction of a documented failure pattern: one hallucinated signal cascades through five ungoverned financial agents, opening $2.6M in unauthorized trades. Based on OWASP ASI08 and Liu et al. (2026). Why runtime governance with formal verification is not optional.
Read more →MAREF Research
The endgame of the GBA AI Agent economy is not 'having the most OPCs' but being the first to build five-layer infrastructure — cognition, governance, accounting, market, and organization layers.
Read more →MAREF Research
Governance capability is replacing model capability as the core moat of the AI Agent industry — cross-validated analysis of policy, technology, and investment based on the Longgang 'LongAiQi' incident
Read more →MAREF Engineering
11 blog posts, 18K+ words, 12 case study files, 15 skill manifests, 9 P0 engineering fixes, and a submit-ready arXiv paper. The full account of MAREF's Q3 brand-building campaign.
Read more →MAREF Engineering
Most 'safety' systems rely on promises. MAREF's governance runs on a state machine with a provable invariant: every transition changes exactly one bit, and HALT is an absorbing state that can't be escaped. Here's the math, and why it matters for production agents.
Read more →MAREF Engineering
MAREF doesn't put a single guardrail between agents and disaster — it puts eight. Inside the pipeline: input controller, file safety guard, clipboard sanitization, 19-class threat detection, a 4-tier decision tree, and an immutable audit layer.
Read more →MAREF Engineering
Self-evolving agents need a brake. MAREF uses a Lyapunov-style stability function to prove — not hope — that defense improves round over round: FNR down 60%, FPR down 66.7%, with automatic saturation detection at round ~175.
Read more →MAREF Engineering
Model Context Protocol is how Claude Code, Cursor, and Windsurf talk to tools. Here's how MAREF plugs into that loop — as both an MCP client and an MCP server — so every tool call runs through the governance layer.
Read more →MAREF Engineering
LangGraph is the most popular graph-based agent orchestration framework. Here's how to put MAREF governance on LangGraph nodes today — over MCP and A2A — and an honest look at where the native sidecar adapter stands.
Read more →MAREF Engineering
Instead of asking you to trust marketing charts, MAREF ships its SLA benchmarks in the repo. Here's every target we assert, the exact command to reproduce them, and what each number means for a production agent.
Read more →MAREF Engineering
MAREF's TLA+ formal verification explained in plain language: 5 theorems (Lyapunov Convergence, HALT Absorbing, Gray Code Transition, Safety Gate Integrity, Red Line Immutability) with code examples and honest limitations.
Read more →MAREF Engineering
Agent skill marketplaces face a supply chain threat worse than npm. Three gates (static scan, sandbox test, human review) are the minimum viable defense against autonomous code execution attacks.
Read more →MAREF Engineering
Step-by-step guide to wrapping CrewAI crews with MAREF governance: subgoal interception, circuit breaker, behavior monitoring, safety gate, and audit trail — without modifying CrewAI internals.
Read more →MAREF Engineering
A 10-dimension governance benchmark comparing MAREF, LangGraph, CrewAI, and AutoGen. Includes reproducible microbenchmarks for overhead, circuit breaker performance, and safety coverage scores.
Read more →MAREF Engineering
LangGraph vs CrewAI vs AutoGen vs MAREF: a buyer's guide to agent governance frameworks. Compare runtime safety, formal verification, state machines, and production readiness.
Read more →MAREF Engineering
Runtime safety for multi-agent AI systems: identity isolation, circuit breakers, drift detection, and blast radius control. Production patterns for preventing agent runaway.
Read more →MAREF Engineering
Model-checked with TLA+, red-blue adversarial evolution, and constitutional safety constraints that enable autonomous AI improvement with monitored convergence — no runaway risk.
Read more →Frankie
After three decades as a steadicam operator on Hollywood sets, I walked away to build safety systems for AI agents. Film sets taught me something Silicon Valley hasn't learned: when things go fast, safety isn't optional.
Read more →MAREF Engineering
Most security tools are 'tested.' MAREF is model-checked — the governance state machine is verified against TLA+ specs of its constitutional red-line invariants, and convergence is monitored with a Lyapunov-style heuristic.
Read more →MAREF Engineering
OWASP published the Agentic Top 10 — the definitive threat model for multi-agent systems. MAREF is the first open-source framework to cover all 10.
Read more →MAREF Engineering
Gartner: 40% of enterprise apps will embed AI agents by 2026. 88.4% of organizations already had an AI agent-related breach (AvePoint). The 7 numbers behind the governance gap — each sourced.
Read more →MAREF Engineering
630M repos, half of new code is AI-written, trust dropped from 77% to 60%. CMU found 6M fake stars. AI code has 1.7x more critical bugs. AI users are 19% slower but think they're 20% faster.
Read more →MAREF Engineering
Multi-agent systems are entering production. Without governance, every agent is an unmonitored vector for data leaks, hallucination propagation, and adversarial exploitation.
Read more →