Comp AI announced a $34 million Series A led by Roo Capital and Grand Ventures on Thursday morning (about $37.5M raised to date). The team is building agents that draft security policies, collect audit evidence, and continuously monitor whether controls still hold after the last SOC 2 snapshot.

CEO Lewis Carhart’s product argument is blunt: a company can pass an audit, then two weeks later ship an AI agent that touches customer data or flips permissions — and the audit was never designed to tell you in real time. Humans still review and approve high-stakes outputs; the Series A is aimed at product expansion plus AI-powered penetration testing across codebases and infrastructure.

The Artificial Intelligence Underwriting Company (AIUC) raised a $40 million Series A led by Ribbit Capital (First Harmonic participating), reported September 16. Its AIUC-1 standard runs 5,000+ adversarial tests across six risk domains and requires quarterly retests; Lloyd’s of London-backed policies put capital behind the certificate — ElevenLabs’ first AIUC-1-backed policy reportedly covered $50M for hallucination loss, data leakage, and faulty tool actions.

KPMG became the first Big Four firm to earn AIUC-1 in August 2026, and the standard is showing up in CSA STAR. Cursor, Lovable, Harvey, and UiPath are among the names already engaging the process. The pitch to procurement: stop relying only on vendor self-assessments when agents can take irreversible actions.

On September 16, TechCrunch unpacked Dario Amodei’s weekend proposal to embed third-party evaluators (METR, Redwood, and peers) inside frontier labs with rights to publish findings without editorial control — Sam Altman said OpenAI would commit as well. Evaluators welcome the access but flag NDAs, short on-site windows, and the risk of becoming vendors on the lab’s terms. Meta, SpaceXAI, and Google DeepMind have not matched the pledge; California’s new SB 813 creates a path for state-recognized independent verification organizations.

Separately this morning, Sen. Mark Warner told Reuters Congress should pass AI safety standards by year-end as a first step — citing agent “swarms” jumping fences into banks, water, or hospitals — while acknowledging Trump-aligned Republicans remain wary of heavy federal drama. He wants proactive protocols, not only after-action disclosure, and left open a future federal AI safety agency.

Bloomberg reported overnight that Chinese-founded agentic startup Manus is lining up roughly $500 million at about a $4 billion valuation — its first round since Beijing forced an unwind of Meta’s planned acquisition. Existing backers include Tencent, HSG, and ZhenFund; terms are still in flux.

Separately this morning, Emulate — a UK startup founded by former Google DeepMind world-model researchers Jack Parker-Holder, Matthew McGill, and Philip Ball — is in talks for about $700 million at a ~$3 billion pre-money valuation to build systems that simulate and predict physical-world behavior. Together the deals underline how fast capital is repricing agent platforms and world-model bets even after geopolitics scrambled one of the year’s biggest AI M&A paths.

Transient, a New York company building a secure operating layer for capital-markets AI agents, secured strategic Series A participation from Nasdaq Ventures alongside lead NEXT Investors, per a September 17 report. The platform sits between agent systems and legacy market infrastructure, applying a Declarative Agentic Framework so policies constrain what agents may do — with real-time oversight, sandboxing, and a claim of zero external data retention.

Nasdaq Ventures’ Gary Offner highlighted embedding governance beside deployment at scale. Transient is also shipping Caddie.AI for trade-lifecycle workflows and plans expansion in London, Singapore, Tokyo, and Hong Kong. The wedge is narrow on purpose: regulated institutions that cannot treat general-purpose agents as a free-for-all against core ledgers.