What are the key security questions organizations should ask when adopting AI tools internally?
Answered at 18:18 in Shadow AI and tokenmaxxing: what your AI habits could cost you with Amy Gebhardt · Random but Memorable
What are the key security questions organizations should ask when adopting AI tools internally?
Organizations should ask: Do you know what credentials are being used for AI tools and workflows and where they are stored? How can you shut down access quickly if needed? What access are you actually giving to AI tools—for example, full file access when only specific files are needed? Understanding boundaries and permissions is critical because giving permission to read files means the AI will read all files, not just the ones intended.
Full answer and detailsRead the whole answer, key facts and every other question from this episode
In short
Organizations should ask: Do you know what credentials are being used for AI tools and workflows and where they are stored?
Amy Gebhardt, Senior Engineering Manager for 1Password SaaS Manager at 1Password · Shadow AI and tokenmaxxing: what your AI habits could cost you with Amy Gebhardt · Random but Memorable · answer at 18:18
Key takeaways
- Organizations should ask: Do you know what credentials are being used for AI tools and workflows and where they are stored?
Quote to remember
“Token maxing is kind of this proxy for productivity or this proof that folks are engaging with AI. Leaders want to measure that their teams are actually embracing and engaging with the technology via the concept of token maxing.”
— Amy Gebhardt
The full answer · about 3 min
Organizations should ask: Do you know what credentials are being used for AI tools and workflows and where they are stored? How can you shut down access quickly if needed? What access are you actually giving to AI tools—for example, full file access when only specific files are needed? Understanding boundaries and permissions is critical because giving permission to read files means the AI will read all files, not just the ones intended.
Key facts
- Unit 42 investigation: attacker used multiple frontier AI agents to work through more than 50 MITRE attack techniques in under 10 hours
- Equivalent manual intrusion activity: approximately two weeks
- Attacker-generated report: 80-page detailed analysis of network weaknesses left on desktop
- Amy Gebhardt's software engineering experience: approximately 20 years
People, companies and topics mentioned
More answers from this episode
- 3:00How did Unit 42 demonstrate the speed increase of AI-assisted cyber attacks compared to manual hacking?
- 5:30How can defenders use AI to detect and respond to machine-speed attacks more quickly?
- 15:00What is the 'living off AI' attack technique and why is it a concern?
- 22:31What is token maxing and why has it become a concern for businesses adopting AI tools?
- 24:39How does AI consumption-based pricing differ from traditional cloud consumption models in terms of predictability?