Enterprise AI's spend-and-breach year opens a door for smaller shops

88.4% of organizations hit at least one agent-related security incident, Uber burned its annual AI budget in four months, and governance delays average nearly six months.

88.4% of organizations experienced at least one agent-related security incident in the past twelve months, up from 75.1% reporting generative-AI breaches the year before, according to the third annual State of AI Report that AvePoint released with Osterman Research on June 29. The 750-respondent survey lands as enterprise AI hits a familiar mid-adoption wall: usage is running ahead of governance, and governance is running ahead of anyone’s ability to price it.

Data leakage led the incident list at 50.1%, with prompt manipulation close behind at 49.6%. Meanwhile the share of organizations that can’t tell whether employees are using unsanctioned AI tools jumped from 6.3% to 17.6% year-over-year. For employee-built agents specifically, the blind spot sits at 21.1%. “Speed of deployment becomes speed of exposure,” said AvePoint CEO Dr. Tianyi Jiang. 86.9% of organizations have delayed generative-AI deployments over data-security concerns; 86% have delayed agent deployments for the same reason. Average delay: nearly six months.

The spend side is stranger. Uber burned through its entire annual AI coding budget in the first four months of 2026, largely on Claude Code, per Fortune. CNBC reports one unnamed enterprise spent $500M in a single month on Claude after issuing employee licenses with no usage caps, and that Microsoft has canceled most internal Claude Code licenses. “Everybody is experiencing the same spend crunch on AI,” said Highspring president Jeff Henry, who notes “countless midsize companies” haven’t started experimenting yet and that some clients are pausing 12 to 18 months before further commitments. Uber COO Andrew MacDonald put the profit question plainly: “the link is not there yet.”

The vendor response is governance retrofitting. Anthropic shipped group- and user-level analytics, model entitlements, and spend alerts for Claude Enterprise on July 2. OpenAI is pitching Frontier as a unified governance layer, disclosing that enterprise now exceeds 40% of revenue and its APIs process 15 billion tokens per minute.

The opening in that gap between hyperscale spend and midsize hesitation is where operators like LemonLime tend to do their best work: right-sized deployments where the visibility question gets answered before the invoice arrives, not after.