The Trust Layer Is Now the Bottleneck for Scaling Agents

AI Agents

The Trust Layer Is Now the Bottleneck for Scaling Agents

What happened: In a single 24-hour window, two companies building the governance and security plumbing for AI agents raised significant capital. Cyera pulled in a $400 million Series G extension from Goldman Sachs at a $12 billion+ valuation to build what it calls the “trust layer for the agentic enterprise” — a platform that governs what data and systems any agent (or non-human identity) can actually reach. Separately, Palma.ai raised $1.8 million in pre-seed to route every agent-to-tool request through a policy engine that can allow, deny, rate-limit, or require human approval at action time.

Why it matters: The story here isn’t the dollar figures — it’s what they signal about where the industry has landed. Eighteen months of “give agents the keys” experimentation has produced a clear consensus: agents are only as safe as the harness around them. Cyera’s CEO put it bluntly: “Global 2000 enterprises need to trust what AI agents can see and do before they can scale them.” The security architecture we have was built for people and applications — not millions of autonomous actors reasoning, calling tools, and moving data at machine speed. Governance is no longer a nice-to-have; it’s the gate in front of every serious production deployment.

What’s next: Watch for consolidation in this layer. Cyera’s acquisition of Oasis Security (non-human identity management) and Palma’s identity integrations with Entra and Okta point the same direction: data access + identity + agent accountability are converging into one category. The companies that win will be the ones that can prove a real audit trail — showing who (human or agent) did what, with what credentials, and why.

Source: Cyera announcement · Source: Palma.ai

Firecrawl Raises $75M to Feed Agents Clean Web Data

Firecrawl closed a $75 million Series B (led by Smash Ventures) and launched Alexandria, a data library that lets agents reach official providers, curated datasets, and the live web through a single API — no custom connector per source. It matters because every agent is only as good as the data it can reliably reach, and “get me clean web data” is quietly one of the biggest hidden costs in agent builds. Source

Claude Opus 5.5 and GPT-6 Sol/Luna Kick Off a Price War — With Benchmarks to Back It

Anthropic shipped Claude Opus 5.5 (20% cheaper, ~40% lower cost on typical workloads) and OpenAI answered hours later with GPT-6 Sol and Luna, slashing API costs 50% or more. The real signal: both vendors are now competing on agentic benchmarks — Opus 5.5 hit 66.4% on Terminal-Bench 4.0 and 40% on AutomationBench — meaning the price war is being fought over who runs your agents cheapest, not just who chats best. Source

AI News

Snorkel AI Triples Valuation to $3.5B as Training Data Becomes the New Moat

Snorkel AI raised $350 million at a $3.5 billion valuation, with revenue up 17–18x since it launched a data-as-a-service offering a year ago. The takeaway: as frontier labs hit diminishing returns on model size, the competition has shifted to who has the best training data — and the pick-and-shovel vendors selling it are now worth billions. Source

California Moves Toward an AI “Kill Switch” for Frontier Models

Gov. Gavin Newsom signed an executive order to “dramatically accelerate” AI oversight, directing experts to recommend within two months whether frontier labs need mandatory third-party safety plans, on-site auditors, and an emergency “kill switch.” With federal action stalled, the “California effect” means these rules could effectively become the national standard — and smaller AI builders may bear the compliance cost most heavily. Source

Zoho’s Vembu: LLMs Are Becoming a Commodity — And That’s the Point

Zoho CEO Sridhar Vembu argued that large language models are rapidly commoditizing, with too many models flooding the market and prices falling fast — and called Big Tech’s push to slow AI development “cynical.” It’s a useful counterweight to the funding headlines: if the underlying models are becoming interchangeable and cheap, the durable value shifts to the applications, data, and trust layers built on top of them. Source