Amazon Web Services (AWS) is embedding its AI-driven security infrastructure within the coding environments of two major competitors, betting that controlling the security layer is more crucial than controlling the AI model itself.
This month, at Black Hat USA 2026, AWS unveiled its Continuum platform, designed to address code vulnerabilities. This platform will integrate seamlessly with Anthropic’s Claude Code and OpenAI’s Codex, in addition to AWS’s Kiro IDE.
This integration brings AWS’s security tools directly to the code-writing stage, independent of the AI model being used. Concurrently, AWS has expanded its Security Hub Extended — a curated, single-bill security marketplace introduced in February — by adding a 10th security category focused on supply chain protection, partnering with Chainguard and Socket.
Collectively, these initiatives represent AWS’s largest effort to establish itself as the primary security control platform for enterprise software development in the AI era. This move has significant commercial implications, especially with the global cloud infrastructure market now exceeding $143 billion per quarter, as reported by Synergy Research Group.
How frontier AI models exacerbated the vulnerability backlog
The urgency for these launches stems from a pivotal moment earlier this year that transformed enterprise security. In April, Anthropic unveiled the Claude Mythos Preview, a general-purpose AI model showcasing unprecedented cybersecurity capabilities.
During pre-release tests, Mythos uncovered thousands of previously unknown zero-day vulnerabilities across major operating systems and web browsers. Alarmingly, over 99% of these vulnerabilities remain unpatched, and the average time from discovery to weaponized exploit is projected to drop to under one hour by the end of 2026.
Chet Kapoor, AWS’s vice president of search, security, and observability, highlighted the challenge in an interview with VentureBeat: “CISOs have long faced code vulnerabilities, but Mythos exacerbated the issue,” Kapoor explained. “The backlog of vulnerabilities has increased fivefold, presenting a significant problem.”
To address this challenge — the rapid growth of known vulnerabilities outpacing organizational capabilities to manage them — AWS developed Continuum. Kapoor described AWS’s broader security vision as transitioning from “telemetry, storage, query, dashboards for humans to telemetry, context, reasoning, and actions by agents.” AWS summarized this vision at Black Hat as “autonomous security at machine speed.”
Exploring Continuum’s four-phase system for automatic code flaw detection and resolution
Continuum operates using what AWS terms an “agent-team loop architecture” — a complex orchestration system that selects the appropriate AI model for each task, connects to the customer’s environment, and delivers validated secure code. It comprises four distinct phases.
Kapoor detailed these phases to VentureBeat. The Discovery phase employs multiple frontier AI models to scan code and incorporate a customer’s existing vulnerability backlog. The Prioritization phase — which Kapoor called “one of our biggest value adds” — contextualizes each finding based on the customer’s environment and business risk. “You go from 100 to 2,000, and then you realize you don’t know which 100 to focus on,” Kapoor noted.
The Validation phase constructs reproducible exploits in a secure sandbox to confirm if a vulnerability is genuinely exploitable. “Once I conduct the tests, how will it behave?” Kapoor asked. “You create a sandbox to explore that. This helps determine the blast radius.” This phase covers both first-party code written by customers and third-party open-source code they rely on. Finally, Remediation provides solutions — such as network configuration changes, policy adjustments, or code patches — that the system has already tested in the sandbox. Human oversight is maintained throughout, allowing organizations to approve outcomes at their preferred level of autonomy.
The commercial model is equally intentional. Customers pay a single price for Continuum, while AWS absorbs the underlying token costs for the most effective frontier model at each phase. “The customer purchases Continuum, period,” Kapoor told VentureBeat. “We determine which model to use for various tasks because, frankly, GPT Cyber excels in some areas, while Mythos excels in others.”
How AWS persuaded OpenAI and Anthropic to integrate a competitor’s security layer
A key strategic element of the announcement is the integration with OpenAI’s Codex and Anthropic’s Claude Code. AWS directly competes with both companies in cloud AI services. Despite a substantial investment in Anthropic and OpenAI’s growing infrastructure competing for similar enterprise AI workloads, both companies agreed to embed Continuum within their developer environments.
In response to a VentureBeat inquiry about competitive dynamics, Kapoor dismissed the notion of competition. “Who is the competitor?” Kapoor asked. “I prefer to view Anthropic and OpenAI as partners. I don’t understand the use of ‘competitor’ in your question.” He added: “They’re partners. We use their models and integrate with their environments, which is why we brought them together for this initiative.”
Kapoor argued that collaboration with a single model provider is insufficient. “It’s not enough to work with just one company,” he said. “Over time, companies will surpass each other.” By absorbing token costs and presenting a single bill to the customer, AWS positions Continuum as infrastructure, not just a model wrapper. The harness, not the engine, becomes the enduring competitive asset.
In his blog post announcing the partnership, Kapoor likened the AI harness to an orchestration layer that connects a model to tools, guardrails, memory, and workflows to deliver outcomes. “Think of the model as the engine and the harness as everything around it. You need both for a high-performance car,” he wrote.
AWS partners echoed this logic. “Model choice was never the hard part for enterprises. Trust in what the model does in production is,” said Val Henderson, CEO of AWS Premier Partner Caylent, as reported by CRN.
AWS expands its curated marketplace to include supply chain security amid rising open-source threats
The second aspect of AWS’s Black Hat announcements is the extension of Security Hub Extended to encompass supply chain security as its 10th category, with Chainguard and Socket as curated partners. The Extended plan now comprises 23 curated partner solutions, all on a single AWS bill without long-term commitments, covering endpoint, identity, email, network, data, browser, cloud, AI, security operations, and now supply chain.
Michael Fuller, AWS’s director of security services, explained to VentureBeat that the addition was entirely customer-driven. “In the last six to eight months, there’s been significant attention on the supply chain space due to widespread reliance on open source,” Fuller said. “Our customers quickly reached out, expressing interest in a supply chain security category with key players because it’s a hot topic for them.”
The two partners were selected to complement, not duplicate, each other. Chainguard provides hardened, secure-by-default container images and packages built from verified source code. Socket monitors package behavior as they’re integrated into a developer’s environment, detecting threats like typosquatting, maintainer account takeovers, and obfuscated malicious code. “Together, we consolidate efforts — Chainguard offers robust, clean images and packages, while Socket conducts behavioral analysis — resulting in a comprehensive supply chain security solution,” Fuller explained.
This complementary strategy addresses two distinct attack vectors. An attacker may release a malicious package containing no known vulnerabilities — Chainguard’s clean-build approach counters this. Alternatively, an attacker could compromise a legitimate maintainer’s account to push a tainted update to a trusted package — Socket’s behavioral detection identifies this. Both vectors are heightened in the AI coding era, Fuller noted, as AI agents face similar supply chain risks as human developers. “Agents can be deceived into downloading a well-known package that has been maliciously obfuscated,” he said.
Why AWS opted for two partners per category instead of building a security marketplace
The partner selection strategy for Security Hub Extended reflects a deliberate philosophy that differentiates it from the AWS Marketplace, which already hosts thousands of security offerings.
Fuller informed VentureBeat that customers clearly articulated their preferences. “They didn’t want hundreds of offerings — we already have the AWS Marketplace,” he said. “They wanted a sweet spot, asking for two options per category, without head-to-head competitors. They wanted one they were familiar with, an established player, and another offering a unique approach.”
Fuller cited the security operations category as an example. “You have Splunk, a well-established leader in security operations, alongside Seven AI, which offers a distinct approach — they complement each other in many ways.”
The decision to build internally versus partner follows a similar rationale. For endpoint detection and response, AWS has no inherent advantage, so it partners exclusively. For cloud security, AWS develops its own native tools due to its deep understanding of its infrastructure, while still partnering with Upwind to provide customers with an alternative option.
“Ultimately, our goal is to ensure our customers can operate securely on AWS, not to grow a large security business as the main objective,” Fuller said. “That’s why we easily choose to build our own solutions while also inviting partners to join.”
The pricing model supports this accessibility. Fuller noted that customers demanded pay-as-you-go options in addition to traditional multi-year commitments. “All Security Hub Extended offerings have a public-facing, pay-as-you-go price, just like our first-party offerings,” he said. “This allows customers to test, scale up, and use the services without a traditional sales cycle.”
The emergence of shadow agents and AI cost harvesting as new cloud security threats
Both AWS executives addressed the growing concern among CISOs regarding the proliferation of unregistered AI agents, termed “shadow agents,” and the novel attack patterns they enable.
Kapoor acknowledged the genuine and increasing issue of shadow agents but clarified that it was separate from the Continuum announcement. “Many agents are registered with directories like Vertex or Agent Core, but numerous agents remain unregistered, earning the name ‘shadow agents,’ which can cause harm,” he explained. “Discovering shadow agents is challenging, but the industry is addressing it.”
Fuller provided more detailed information on AWS’s current deployments. Security Hub now offers a free AI inventory capability utilizing three data layers: AWS Config identifies AI-related services like SageMaker, Bedrock, and Agent Core across an organization; Amazon Inspector scans compute instances and containers for AI-related software; and GuardDuty analyzes DNS request and response logs for known AI tools and agentic workloads.
Beyond inventory, Fuller revealed that GuardDuty now monitors data plane events, including prompts, prompt volume patterns, and inference cost analysis, to detect what AWS calls “cost harvesting.”
This attack mimics the cryptocurrency mining that became prevalent after cloud credential compromises: an attacker gains access to an AWS account and utilizes as much free AI inference as possible before being detected.
“We’re observing what we term cost harvesting,” Fuller stated. “Attackers will spin up operations to maximize free inference usage until discovered.” Fuller noted that “similar activities are occurring in AI” as they did in crypto mining, and GuardDuty’s detection of credential compromise and unauthorized compute usage directly translates to this new threat.
Integrating Continuum and Security Hub Extended within AWS’s enterprise security strategy
Although the announcements were made in the same week, AWS treats the products as separate entities. Kapoor described Continuum as distinct from Security Hub Extended, each sold as standalone products. AWS did not disclose its long-term plans for the two.
The design logic points in a specific direction. Continuum focuses on the code an enterprise writes and the open-source code it inherits. Security Hub Extended addresses everything else, with the newest category being where the two clearly overlap. Continuum’s validation phase covers third-party dependencies alongside a customer’s code, while Chainguard and Socket fortify and monitor the same packages from another angle. One capability is developed in-house, the other curated from partners, and they meet at the same attack surface.
Both initiatives stem from the belief that enterprises prefer recommendations over catalogs.
“Customers desire an informed perspective on how to approach security in the AI era,” Kapoor told VentureBeat. “That’s the purpose of Security Hub Extended — providing our opinion.” AWS will continue offering customers choices, he added, “whether through our products or partner offerings.”
This philosophy — a recommendation with flexibility — connects a curated marketplace to a first-party agent platform and blurs the line between them more than the separate announcements suggest. Security Hub has already integrated capabilities that didn’t exist a year ago, including the free AI inventory and the cost harvesting detections Fuller described. The console serves as AWS’s venue to deliver its perspective to enterprises. Continuum is the strongest opinion it has introduced.
The audience for these recommendations has evolved, Kapoor noted. Mythos, he argued, shifted security from a CISO responsibility to a CEO and board-level priority. “Boards now request updates on enterprise security because it poses a business threat and risk.”
AWS’s security ambitions represent a strategic bet on controlling the orchestration layer
The dual launches are part of a broader strategic expansion AWS has pursued throughout 2026 at a rapid pace. The company revamped Security Hub at re:Invent 2025 by consolidating GuardDuty, Inspector, CSPM, and Access Analyzer into a single console. In February, it launched Security Hub Extended with 14 curated partner solutions, which grew to 21 across nine categories by May. It now includes 23 across 10 categories. Continuum debuted at the New York Summit in June and extended to OpenAI and Anthropic integrations at Black Hat in August.
AWS reported $42.2 billion in revenue during Q2 2026, with cloud sales growing 37% year over year. The company holds a 28% share of the global cloud infrastructure market, outpacing Microsoft at 20% and Google at 15%.
Fuller told VentureBeat that AWS has “tens of thousands of customers using one or multiple security services across all regions and industries, including commercial and government sectors.” The Extended plan seeks to convert this installed base into users of partner security solutions, deepening engagement and making it harder for competitors to displace AWS as the default platform.
By becoming the seller of record for 23 partner security solutions and embedding Continuum within the coding environments of OpenAI and Anthropic, AWS is constructing something more enduring than a product line. It is creating the connective tissue between enterprises and every AI model they use, every open-source package they integrate, and every security vendor they employ. In a rapidly advancing world of frontier models, where today’s best scanner quickly becomes obsolete, the enduring layer is not the model itself but the harness connecting the model to the customer’s environment, policies, and risk tolerance.
Reflecting on a conversation during a flight to Black Hat, Kapoor shared a straightforward reason for the importance of these efforts. A CISO turned CTO sitting next to him expressed a candid view of the current state: “I don’t feel safer now.” Kapoor’s response, as he told VentureBeat, was equally direct: “We’re working on it.”
Whether this work ultimately enhances global safety or makes AWS indispensable to organizations seeking that safety, the outcome may be the same.
