TUI's Platform and Engineering security team had four engineers supporting more than 100 digital product teams, and could complete only three to four threat-modeling sessions a month before the new system, according to an AWS blog post describing TUI's work. [1]
TUI built its system on Amazon Bedrock using Anthropic's Claude Sonnet 4.5, developing it during a 10-week proof of concept with the AWS Generative AI Innovation Center. The post says TUI moved the system into production within 30 days of handover. [1]
The team's first approach loaded all 192 internal security standards, about 338 KB, into the model's context and ran into token limits. The team then switched to metadata-driven selection, in which the model picks 16 relevant standards, reducing context to about 21 KB. [1]
TUI reports that time per assessment fell by up to 70% from four to six hours, that monthly sessions rose from three to four to 10 to 12, and that the system identifies 40 to 60 threats per architecture compared with 30 to 40 in manual sessions. The post notes these figures come from TUI's internal measurements comparing before and after deployment. [1]
The caveat is that these results are company-reported and appear in a vendor-published case study, so they should be read as TUI's account rather than independent confirmation. [1]
To limit repeated external lookups, the system caches web search results in Amazon S3 with a seven-day time-to-live. TUI's monitoring reportedly shows a 70% to 80% cache hit rate, since many of its applications share core technologies. [1]
Marvin Jordan, Head of Technology for Security Platform and Practices at TUI, said that in the first four weeks after launch 64 users registered and 34 threat models were produced. He said adoption had grown to 255 registered users, 86 threat models and 43 active creators as of October 2026. [1]
According to the post, architecture diagrams and threat models are processed in the AWS eu-central-1 Frankfurt Region, encrypted at rest with AWS KMS and in transit with TLS. [1]
TUI plans to add automated Jira ticket creation for mitigation tracking and integration with CI/CD pipelines for pre-deployment security checks. [1]