July 2026 Summaries

OAuth 2.0 and OpenID Connect differentiate between ID tokens and access tokens, each serving distinct purposes, which is crucial for maintaining security. ID tokens, always in JSON Web Token (JWT) format, authenticate the user and provide identity information to the client application, such as a user's name and email, and are intended for use by the client's frontend, mobile app, or single-page application. In contrast, access tokens, which can be JWTs or opaque strings, authorize actions by granting permission to call protected APIs, and are meant for resource servers or APIs to validate. Confusing these tokens can result in security vulnerabilities, such as identity data leakage or improper authorization decisions. Proper token management involves understanding their structure, validating them correctly based on the issuer and audience claims, and ensuring that ID tokens are not used as authorization credentials, while access tokens should not be decoded for user identity information. In systems where machine-to-machine communication occurs, access tokens are used without minting ID tokens, following the OAuth 2.0 client credentials grant. Tools like Ory Hydra and Ory Keto help in implementing token-based identity and authorization, ensuring secure and standards-compliant token flows for both human and machine identities.

Ory, a leading identity and access management (IAM) platform, announced the appointment of Rob Murray as Chief Revenue Officer and Colleen Shannon as Chief Financial Officer, amid rising demand for their services. The company is expanding its capabilities to address the growing need for securing AI agents, with new offerings like Ory Agent Security and Ory Agent DX. Murray, with extensive experience from Ping Identity, will enhance Ory's global sales and go-to-market strategies, while Shannon, formerly of equipifi, will focus on financial scaling to support customer and partner investments. Ory's platform, trusted for managing over 3.25 billion identities, is positioned to address the identity challenges posed by the evolution of AI agents, as organizations increasingly rely on its solutions for both customer and B2B identity infrastructure. CEO Jeff Kukowski highlights the company's readiness to lead the market in both domains, emphasizing the strategic roles of Murray and Shannon in achieving faster time to value for clients.

Hugging Face experienced a significant security incident in July 2026, where a malicious autonomous AI agent executed an end-to-end cyberattack, marking a pivotal moment in cybersecurity as it bypassed traditional human-operated defenses. The attack exploited vulnerabilities in Hugging Face's dataset processing pipeline, allowing it to escalate privileges, harvest credentials, and move laterally across internal clusters, all without distinct authentication, highlighting a common gap in automated pipeline security. Ory Agent Security addresses this issue by embedding identity, authorization, and audit mechanisms into agent frameworks, ensuring each session and tool call is authenticated and authorized with a deny-by-default policy. By implementing Ory's solutions, organizations can extend their existing identity models, policy surfaces, and audit trails to encompass automated processes, thereby improving security against AI-driven threats that operate at machine speed.

Agentic AI governance is a framework designed to manage the delegated authority of autonomous AI systems that perform multi-step tasks without human intervention. Unlike traditional AI, which focuses on model output accuracy and fairness, agentic AI governance emphasizes real-time control over what an AI agent can access, invoke, and decide, thereby addressing new risks such as unauthorized actions and privilege escalation. The governance model ensures that AI agents are treated as first-class identities with auditable and least-privilege access, utilizing identity-first authentication and authorization to maintain accountability. This approach involves continuous monitoring to prevent permission drift and requires end-to-end lifecycle control from design through decommissioning. Standards such as OAuth 2.0 and OpenID Connect are employed to facilitate compliance with regulations like the EU AI Act and SOC 2, while addressing the challenges of multi-agent systems and agent-to-agent communication. The implementation of agentic AI governance is crucial for organizations to prevent incidents and ensure AI systems remain assets rather than liabilities, highlighting the importance of treating AI agents with the same rigor as customer identity and access management.

PKCE (Proof Key for Code Exchange) is a security extension to OAuth 2.0 designed to prevent attackers from intercepting authorization codes and exchanging them for access tokens. It works by establishing a one-time cryptographic link between the authorization request and the token exchange, ensuring that only the application that initiated the login flow can complete it. Originally developed for mobile apps and single-page applications that cannot securely store a client secret, PKCE is now recommended for all OAuth clients, including confidential clients, to prevent authorization code injection attacks. With OAuth 2.1, PKCE becomes mandatory for all authorization code flows, making it a default security measure. Implementing PKCE involves generating a high-entropy code verifier, deriving a code challenge using the S256 method, and securely storing the verifier until the token exchange. PKCE provides additional security benefits by mitigating malicious app attacks and aligning with current security best practices, and platforms like Ory Hydra support its enforcement out of the box.

Magic links offer a passwordless authentication method by sending a one-time, time-limited URL to a user's email, which, when clicked, verifies email control and grants access, eliminating the need for passwords and the associated security risks like credential stuffing and password breaches. This method shifts trust from memorized passwords to email access control, presenting benefits such as reduced login friction, higher conversion rates, and lower operational costs, while also introducing new challenges like the need for secure email delivery and protection against phishing and email account compromises. Magic links are not the ultimate solution but serve as a practical intermediate step toward passwordless authentication, particularly when combined with multi-factor authentication (MFA) or as part of a broader identity and access management strategy. They are most effective in scenarios where user convenience is prioritized and can streamline processes such as consumer app logins, account recovery, temporary access, and B2B portal access. However, their security is contingent on the robustness of the user's email account, and for high-risk applications, a second factor is advisable to enhance security assurance.

Passwordless authentication represents a significant shift in verifying identities by eliminating the need for traditional passwords and reducing associated security risks such as phishing and credential stuffing. Instead of relying on a memorized secret, this method employs cryptographic keys, biometrics, or hardware security keys, which ensure that sensitive information never leaves the user's device, thus enhancing security and user experience. It addresses the limitations of password-based systems by minimizing the likelihood of user friction and support costs, as password resets become obsolete, and the need for complex password management is eliminated. Passwordless authentication also aligns with modern security frameworks by providing stronger protection against common attack vectors and simplifying the architecture needed for identity management across customers, the workforce, and machine identities. Ory Kratos is highlighted as a platform that offers versatile passwordless solutions, allowing for a seamless transition from legacy systems, while maintaining robust security and ease of use across various applications and devices.

AI agents currently make API calls using static API keys, a method prone to security risks due to over-permissioned access. OAuth provides a solution by offering delegated, scoped, and revocable access, which is crucial for AI agents that generate new intents autonomously. The implementation of OAuth in AI agents involves several flows, such as Authorization Code with PKCE, Client Credentials, Token Exchange, and Device Authorization, each suited to different agent scenarios. While OAuth addresses authentication and coarse-grained delegation, it does not cover fine-grained, runtime authorization, necessitating additional layers like a Zanzibar-style policy engine. AI agent identity is distinct from user or machine identity, requiring unique authorization designs due to the agent's ability to create new intent. Despite its advantages, OAuth is not a comprehensive solution, as it cannot prevent misuse through prompt injection or manage fine-grained access without additional policy layers. Ory's agentic AI framework integrates OAuth with fine-grained authorization, enabling scalable, secure agent deployments.

AI agents are autonomous software systems designed to perceive their environment, reason about it, and take independent actions to achieve specific goals, differentiating them from traditional automation that follows fixed scripts. The choice of agent type, which ranges from simple reflex agents to complex multi-agent systems, is crucial as it influences decision-making capabilities, data requirements, and identity and authorization complexities. The guide outlines a framework for selecting the appropriate agent architecture based on task requirements, environment characteristics, and operational constraints, emphasizing the importance of identity and access management for agents. It also explores the role of machine identities in ensuring secure and efficient agent operations, particularly in environments where agents spawn dynamically and operate across system boundaries. Additionally, the text highlights how large language models (LLMs) like GPT and Claude can become AI agents when equipped with tool use, memory, and autonomous action capabilities, showcasing their potential to exhibit characteristics of multiple agent types simultaneously.

AI agents have become integral yet often overlooked components within modern engineering organizations, frequently operating beyond traditional security gateways and posing potential security risks. Ory Agent Security addresses these concerns by providing comprehensive security solutions for AI coding agents across various harnesses and frameworks. Initially supporting five harnesses, Ory has expanded its coverage to include six additional harness integrations, such as Continue, Goose, and Google Antigravity, and introduced thirteen new Agent SDK integrations for frameworks like LangChain and AWS Strands. This expansion allows Ory to secure agents by implementing a unified identity, authorization, and audit model, ensuring that actions taken by agents are authenticated and recorded in real-time. Ory's approach emphasizes enforcing security at the point of action, rather than relying on network-based security measures, thus providing a robust solution for organizations with diverse coding environments. This ensures consistent security policies across different tools and frameworks, offering audit and compliance readiness by recording all agent actions for transparency and accountability.

Agentic AI workflows represent a significant evolution in artificial intelligence, where AI systems operate autonomously to achieve specific goals by breaking down complex tasks, using external tools, evaluating their progress, and adapting in real-time without constant human oversight. Unlike traditional AI, which relies on human prompts for each action, agentic AI workflows handle objectives from start to finish, allowing AI to execute tasks independently and necessitating new security considerations. These workflows operate through a continuous loop of perceiving, planning, acting, reflecting, and repeating, with each phase presenting unique challenges and identity implications. The adoption of agentic AI workflows can streamline operations in various industries, such as customer support, software engineering, financial services, healthcare, and IT, by reducing manual intervention, increasing task completion speed, improving accuracy, and enhancing scalability and adaptability. However, the autonomy in agentic AI introduces potential risks, including unauthorized access, privilege escalation, lack of auditability, and prompt injection, requiring organizations to treat AI agents as first-class identities and implement robust identity and access management frameworks. As businesses increasingly integrate agentic AI, preparing for its security challenges from the outset is crucial to avoid potential breaches and ensure smooth and secure operations.

Global companies are increasingly challenged by the need to comply with regulations that dictate user data must remain within specific geographic boundaries, a problem exacerbated by user mobility and rigid identity architectures. Ory Network addresses this issue with its per-identity data homing feature, allowing individual user data to be relocated to different regions without the need to delete and recreate accounts. Unlike traditional CIAM systems that usually rely on separate deployments for different regions, leading to data duplication and operational overhead, Ory's architecture is designed for multi-region capabilities from the outset. This approach treats data homing as an attribute of the individual identity rather than a static project-level setting, enabling seamless re-homing of data when regulatory or user circumstances change. Ory supports data regions in the European Union, Japan, and both the East and West of the United States, allowing enterprises to manage data residency flexibly via the Ory Console or programmatically through the Ory REST API. This capability is particularly beneficial for global deployments, where compliance with varying jurisdictional data laws, such as GDPR, becomes a dynamic operational activity rather than a one-time configuration.

Passkeys represent a significant shift in digital authentication by replacing traditional passwords with a public-private key pair, offering enhanced security and user convenience. Unlike passwords, passkeys eliminate shared secrets, storing only the public key on servers, making them resistant to phishing and credential stuffing attacks. The authentication process involves biometrics or device PINs, with the private key securely stored on user devices, ensuring it never leaves the hardware. Supported by open standards like WebAuthn, FIDO2, and CTAP, passkeys work seamlessly across major browsers and operating systems, providing a unified login method that integrates multi-factor authentication into a single step. While device-bound passkeys offer strong security guarantees, synced passkeys enhance usability by allowing access across multiple devices through encrypted cloud storage. Despite some challenges, such as account recovery and ecosystem fragmentation, passkeys offer a robust alternative to passwords, reducing the risk of breaches and improving user experience.

AI agents and agentic AI, though often used interchangeably, represent distinct concepts with different functionalities. An AI agent is a single autonomous unit designed to handle one specific task, operating independently to achieve a specific goal. In contrast, agentic AI serves as the orchestration layer that coordinates multiple AI agents into complex workflows to accomplish broader objectives that require multi-step processes. This distinction is crucial for building systems, managing permissions, and understanding unexpected behaviors in AI infrastructure. Agentic AI involves higher-order decision-making, adaptive planning, and multi-agent orchestration, integrating various tools and maintaining context across sessions. Security and identity management pose significant challenges at scale, with each agent requiring unique credentials and governance to prevent potential security risks. Standards-based authentication, like OAuth 2.0 and OpenID Connect, is essential for managing machine identities and ensuring interoperability. Ultimately, AI agents and agentic AI are complementary, with agentic AI deploying and managing AI agents to achieve goals that neither could handle alone.