Technology & SaaS2026
Source: Cloudflare ·
When an AI response contains another customer’s context
Cloudflare disclosed a cross-request context exposure affecting one Workers AI model.
Read the briefingFilter briefings
Find a case like yours
Start with your industry, or explore the access decision you face.
Years refer to incidents or evaluations, not later source reports. Cases without a stated incident year are grouped separately. Filters can be combined.
All briefings
Page 1 of 5
Technology & SaaS2026
Source: Vercel ·
A third-party AI tool became a route into connected systems
Vercel traced an intrusion to a compromised AI tool used by an employee.
Read the briefingTechnology & SaaS2026
Source: Anthropic ·
When drafting becomes an action in the real world
Evaluation cases show why reading, drafting, and submitting need separate permissions.
Read the briefingTechnology & SaaS2026
Source: Hugging Face ·
A processing pipeline became a path into other systems
Hugging Face’s technical timeline traces an agent’s movement from processing infrastructure to credentials.
Read the briefingTechnology & SaaS2025
Source: Microsoft ·
An email should not decide what an assistant reveals
The now-fixed EchoLeak flaw illustrates the risk of treating retrieved content as instructions.
Read the briefingTechnology & SaaS2025
Source: OpenAI ·
Sensitive context includes the metadata around a product
A vendor breach exposed analytics information rather than conversations or API keys.
Read the briefingConsumer services · Technology & SaaS2023
Source: Information Commissioner’s Office ·
Capita: one supplier breach reached millions of people’s records
A supplier attack exposed pension, staff and customer information across client organizations.
Read the briefingGovernment & public sector2021 / 2022
Source: UK Cabinet Office ·
Norfolk and Suffolk police: crime statistics carried hidden personal data
Files released for crime-statistics requests contained hidden information about identifiable people.
Read the briefingNo briefings match these filters
Try a broader search or remove a filter to explore related cases.
Why we’re building Vivendur
These briefings make the need for clearer access boundaries concrete. Vivendur is building private AI memory so organizations can choose what each tool receives for a task. The cases inform that work: limiting unnecessary disclosure, separating permissions between tools, and being clear about what happens after information is shared.
Our approach to reporting
We link to original disclosures, distinguish incident dates from publication dates, and label our interpretation. These cases raise different security questions; they do not establish that Vivendur could have prevented them.
See these access decisions in Vivendur in action