95% of enterprise AI projects fail

AI Business Risk Weekly

This week: MIT reports that almost all enterprise AI deployments are floundering, a $100M lobbying machine emerges to blunt regulation, Otter.ai faces a lawsuit over recording and training consent, Stanford finds early evidence AI is reshaping entry-level hiring, and Grok chats leak publicly through search indexing. Each story points to the growing tension between rapid adoption and the slower-moving safeguards of governance, training, and trust.

95% of Enterprise AI Deployments Fail

A new report from MIT’s NANDA initiative found that just 5% of enterprise AI projects produce revenue. Despite mass enthusiasm, most efforts stumble on poor integration, lack of training, and misaligned expectations- not on technical limits. Many firms launch AI pilots without mapping where tools genuinely augment workflows, leading to shadow adoption by employees and fragmented strategies that mirror the dot-com bubble’s hype cycle. The result is a striking paradox: near-universal employee adoption, but with minimal guidance on risks, ethics, or effective usage.

Business Risk Perspective: The velocity of AI launches has outpaced the slower work of strategy and governance, leaving most projects stranded. Failures are less about capability and more about organizational readiness to embed tools responsibly.

AI Industry Launches $100M Political Influence Network

Andreessen Horowitz, OpenAI’s Greg Brockman, Perplexity, and other investors have launched Leading the Future, a super PAC network modeled after crypto lobbying campaigns, starting with a $100M war chest. The network plans to aggressively back candidates who support a “pro-AI agenda” and oppose those who don’t, aiming to tilt state and federal regulation in favor of industry. With ties to prior efforts that ran misleading ads and used intimidation tactics, this marks a new escalation in how AI firms will try to shape public policy.

Business Risk Perspective: The rapid build-out of industry-funded lobbying raises uncertainty about the durability of emerging safeguards. Regulatory direction may swing with political cycles, complicating long-term compliance planning.

Otter.ai Sued Over Training Consent

Otter.ai is facing a California class-action lawsuit alleging its Otter Notetaker recorded and used meeting conversations for AI training without obtaining proper consent. The case, seeking over $5M in damages, points directly to the complexity of consent laws: some jurisdictions require all parties to agree before a recording is made, while others permit single-party consent. With AI note-taking tools now common in workplaces, many employees and clients remain unaware their words may be stored and repurposed, creating a training gap as well as a compliance gap.

Business Risk Perspective: Consent laws vary dramatically across states and countries, and AI tools often operate across those borders without clear disclosure. Without explicit training for employees, routine use of such tools can unintentionally trigger costly legal exposure.

Stanford Report: AI Depressing Entry-Level Hiring

Stanford’s “Canaries in the Coalmine” study identified an early labor market shift: entry-level job postings are declining in roles like coding where AI automates tasks, while demand for more experienced roles holds steady. This suggests AI is not eliminating work outright but reshaping pipelines, reducing opportunities for juniors to gain skills and advance. If this pattern accelerates, industries may face gaps in long-term talent development, as career ladders thin at the bottom even while mid-level demand persists.

Business Risk Perspective: AI-driven productivity gains may undercut the traditional entry-level path, leaving firms with weaker pipelines of trained professionals. Workforce planning will need to anticipate shortages years downstream, not just short-term cost savings.

Hundreds of Thousands of Grok Chats Indexed by Google

Forbes revealed that conversations shared from Grok generate URLs which are indexed by Google, Bing, and DuckDuckGo, making them searchable by anyone. Users who thought they were privately sharing a conversation with a colleague or friend may have inadvertently exposed sensitive material to the open web. This comes just weeks after a similar incident involving ChatGPT, pointing to a recurring blind spot in how AI platforms handle link-sharing features.

Business Risk Perspective: The false sense of privacy when sharing AI chats creates openings for accidental disclosure of business secrets. Training employees on these risks is essential, as technical safeguards alone are proving insufficient.

AI Business Risk Weekly is a Conformance AI publication.  

Conformance AI ensures your AI deployments remain safe, trustworthy, and aligned with your organizational values.