Developing
The article's conclusion: agent security comes down to what an agent can access, who authorised it, what it did and whether you can prove it.
In 60 seconds
- ASOS said on October 6, 2026 that it was investigating a cyber incident after customers received an unauthorised push notification; names and contact details may have been accessed, but it did not believe payment cards or passwords were affected.
- IBM and Red Hat found and fixed more than 400 previously unknown vulnerabilities in widely used Java libraries, with fixes backported.
- Microsoft made Execution Containers generally available to limit the files and networks AI agents can reach.
- The UK NCSC advised ASOS customers to assume they were affected and be wary of suspicious links.
Impact Layer
- What changed
- Retail CX teams face both customer data incidents and a new wave of agent-security controls.
- Why it matters
- Customer-facing agents that can read data and act need boundaries and audit trails, and a breach shows what is at stake.
- Who is affected
- CX, security and e-commerce teams, especially in retail.
- Martech stack area
- Commerce, Customer Experience
- Stack impact
- Review which data customer-facing agents can access, and prepare customer communications for incidents involving push notifications.
- Business impact
- Moderate A weekly roundup by CX Today. Anthropic's vulnerability counts and Cloudflare's agent design are reported there without independent verification.
- Related technologies
- What to watch
- ASOS's investigation findings and Microsoft's Entra work to separate agent from user activity.
CX Today’s roundup for the week of October 6 to 10, 2026 connects several security stories to customer experience.
ASOS
ASOS confirmed on October 6 that it was investigating a cyber incident after customers received an unauthorised push notification. Basic personal information, including names and contact details, may have been accessed; ASOS said it did not believe payment card information or account passwords were affected. The UK’s National Cyber Security Centre advised customers to assume they were affected and to be wary of suspicious links in push notifications, emails and messages.
Agent security news
- IBM and Red Hat: found and fixed more than 400 previously unknown vulnerabilities in widely used Java libraries, with backported fixes, and made the Lightwell Clearinghouse generally available. They warned that autonomous AI agents could chain several lower-risk flaws into a serious attack.
- Microsoft: Execution Containers are generally available, limiting the files, networks and other resources AI agents can access, with policy kept outside the agent. Through Entra, Microsoft is working to tell agent activity from user activity.
- Cloudflare: its managed-defense team uses specialised agents; an early single-agent prototype produced unsupported claims, so evidence collection was separated from AI analysis, with a human making the final decision.
Our read
The article’s framing is useful: for any customer-facing agent, ask what it can access, who authorised it, what it did and whether you can prove it afterwards.
Sources
- CX Today: What ASOS, AI agents and 400+ vulnerabilities reveal about CX risk (opens in a new tab)
Source of all facts in this article.