Three engineers used ChatGPT within 20 days of Samsung enabling it. One pasted semiconductor source code to fix a bug. One submitted confidential chip yield data for optimisation. One recorded an internal meeting and asked ChatGPT to generate the minutes. All three were working exactly as intended. The data entered OpenAI’s training corpus. Samsung banned ChatGPT company-wide within weeks.
“An agent can behave correctly and still transmit your most sensitive data outside your organisation — with no outbound control in place.”
Egress scanning detects live outbound connections to AI API endpoints. CIM contracts restrict which data domains each agent is authorised to reach.
Source: Bloomberg, Cybersecurity Dive, April 2023.
Could this happen in your organisation?
Find every AI agent operating across your estate in under 10 minutes.
Talk to us