In a column on vc.ru, Vlad Karmakov, founder and CEO of IT company Siberian.pro, breaks down how employees bypass their employers using personal AI subscriptions. He pulls together three numbers: 78% of people drag their own AI into work (BYOAI, per Microsoft and LinkedIn), 39.7% of what gets sent to third-party chatbots is confidential data (Cyberhaven), plus an extra $670,000 tacked onto breach remediation costs when shadow AI is involved (IBM). And separately — the risk of fines under Russia's Federal Law 152.
The mechanics are simple, and that's what makes them ugly. A sales rep buys a personal ChatGPT subscription through a reseller, flips on a VPN, and uploads call transcripts and client correspondence. A marketer spends six months tuning system prompts to match the company's voice — and keeps them in personal notes. A developer runs scripts from a personal email account. While the person is on payroll, everything hums along. The day they hand in their notice, the tools, the data, and the know-how walk out with them.
The author proposes a single secure perimeter built on five principles. Prompts, skills, and agents belong to the company: a successful workflow gets packaged as a corporate Skill — a deal-scoring template for the CRM, a resume screener for HR, a marketing metrics analyzer. Someone leaves? Access is cut with one click via SSO, and the work stays behind. Then there's role-based access: a sales intern physically can't upload a financial report into the assistant. The author backs this up with his own track record: 60+ workstations and platform pilots at 10+ external organizations.
The 78% figure is believable precisely because companies have barely pushed back. The business wants speed, IT delivers a policy six months later, and the employee doesn't wait — they just buy a subscription. Our read: this isn't a technology story. It's the story of centralized AI arriving at the company later than the chaos does.
What's actually changing here is the conversation about money. The $670,000 from IBM's report hits a budget harder than any sermon about security culture. But we'd bet on a different argument: the company isn't just losing data — it's losing accumulated knowledge. One person leaving wipes out months of prompt tuning, and the new hire starts from zero. Security can be fixed with a policy; learned know-how only survives in shared infrastructure.
Who wins? Enterprise platform vendors — expect their op-eds to get more frequent. The losers are open personal accounts and, to some extent, the employees themselves: less freedom, more reporting. In our experience, a corporate alternative only sticks when it's more convenient than the personal one — not merely sanctioned.
All three figures are secondhand citations in a vendor's column — there are no links to the original Microsoft, LinkedIn, Cyberhaven, or IBM reports in the text. There's no breakdown of any specific incident with an actual damage figure either. How much such a perimeter costs and how long deployment takes? Not a word. The author is selling a platform, so read his conclusions as the position of an interested party.
Ban personal ChatGPT on work laptops — or hand out a corporate one from day one and stop getting in people's way?
Is the $670,000 a real budget argument, or a convenient scare tactic for selling a platform?
Prompts someone spent six months polishing: do they belong to the company, or to the person who wrote them?
Forecast: within six months, at least one major Russian systems integrator will launch an "enterprise AI perimeter" product with SSO and role-based access — riding the wave of Federal Law 152.
Disclaimer: the figures and examples come from a single author column and are not officially verified.
Source: vc.ru
Source: vc.ru · post in Telegram