AWS Continuum Secures OpenAI Codex and Anthropic Claude Code
AWS brings Continuum to OpenAI Codex and Anthropic Claude Code, expands supply chain security, and targets AI flaws, shadow agents and inference theft.
Summary
At Black Hat USA 2026, reported August 10, Amazon Web Services said Continuum will integrate with Anthropic’s Claude Code, OpenAI’s Codex and AWS’s Kiro IDE, embedding vulnerability controls where developers work. Its agent-team loop discovers flaws with multiple frontier models, prioritizes them against business context, validates reproducible exploits for first-party and open source code in an isolated sandbox, then offers tested configuration, policy or code fixes subject to human approval. Customers pay one Continuum price; AWS absorbs token costs and selects models, including GPT Cyber or Claude Mythos, by task.
Anthropic’s Claude Mythos Preview, announced in April, found thousands of unknown zero-day flaws across major operating systems and browsers; more than 99% remain unpatched. Median discovery to weaponization fell from 771 days in 2018 to under four hours in 2024 and is projected below one hour by year-end 2026; AWS vice president Chet Kapoor said existing backlogs grew fivefold.
Security Hub Extended, launched in February with 14 curated solutions and expanded to 21 across nine categories by May, now has 23 across 10, adding supply chain security with Chainguard’s verified, hardened images and Socket’s behavioral package monitoring. The single-bill, pay-as-you-go catalog requires no long-term commitment. Director Michael Fuller said free AI inventory uses AWS Config, Amazon Inspector and GuardDuty, which also detects inference “cost harvesting,” while unregistered shadow agents remain difficult to discover. AWS posted $42.2 billion in Q2 2026 revenue, up 37%, and holds 28% of cloud infrastructure versus Microsoft’s 20% and Google’s 15% in a market above $143 billion quarterly. Continuum remains separate from Security Hub Extended, with no combined roadmap disclosed.
Positives
- Continuum integrates directly with Claude Code, Codex and Kiro IDE while preserving human approval over proposed security fixes.
- Four automated phases discover, prioritize, validate and remediate vulnerabilities using tested exploits and patches inside isolated sandboxes.
- One Continuum price covers whichever frontier model AWS selects, with AWS absorbing underlying token costs.
- Security Hub Extended now offers 23 curated solutions across 10 categories on one AWS bill without mandatory long-term commitments.
- Chainguard hardens verified packages while Socket monitors package behavior for typosquatting, compromised maintainers and obfuscated malicious code.
- Free AI inventory combines AWS Config, Amazon Inspector and GuardDuty data across services, instances, containers and DNS activity.
Risks & concerns
- More than 99% of thousands of zero-day vulnerabilities found by Claude Mythos Preview remained unpatched when reported.
- Under one hour is the projected median time from vulnerability discovery to weaponized exploitation by the end of 2026.
- Fivefold vulnerability backlog growth followed Mythos testing, according to AWS vice president Chet Kapoor.
- Shadow agents operating outside registration directories can cause harm and remain difficult for the industry to discover.
- Cost harvesting lets attackers exploit compromised AWS credentials to consume AI inference before account owners detect the abuse.
- AWS declined to provide a longer-term roadmap connecting standalone Continuum with Security Hub Extended.