Smallstep

Everyone is talking about AI security in terms of model safety, prompt injection, and data governance. Those are important conversations. But there's a more fundamental problem that's getting overlooked: LLMs, agents, and workloads are increasingly being given broad, long-lived access to APIs, tools, and data. There's no MFA prompt for an agent. No human approving every action. Once an API key or token is issued, it's often trusted indefinitely, with little visibility into what is actually executing requests. We've seen this firsthand while dogfooding AI internally at Smallstep. As we built more agent-driven workflows, we realized the same identity problems we've solved for users and devices are now emerging for AI. Long-lived secrets don't scale, and they don't belong in autonomous systems. On June 30, join us for a live webinar on Rethinking MFA for Non-Human Identities. We'll share: • What we've learned from securing our own AI workflows • Why traditional MFA doesn't work for agents, workloads, and APIs • How to move beyond static API keys toward cryptographically verifiable identity • A live demo of what "MFA for AI" looks like in practice It's free, it's 30 minutes, and if you're building or securing AI systems, you'll leave thinking differently about authentication. 🔒 Register here! 👉 https://hubs.ly/Q04mGNf30

Your devices have had hardware security built in for years and most companies aren't even using it.😅 Here's something most security teams don't think about: the hardware co-processors that make credential exfiltration impossible have been shipping inside enterprise devices for years. ⏰ TPMs have been standard on Windows and Linux machines since 2016. Apple's Secure Enclave has been in every Mac since the T1 chip. These components were designed to store cryptographic keys in a way that makes them physically impossible to export since the private key is generated on the hardware and never leaves it. The capability has always been there. The tooling to actually use it for device identity, at scale, across an enterprise fleet hasn't been, until now. When Smallstep issues a certificate using ACME Device Attestation, the private key is bound to that specific TPM or Secure Enclave at issuance. You can't copy it to another machine. You can't exfiltrate it. The certificate proves not just that a device is enrolled, but that this exact piece of hardware is making this exact request. Most organizations are sitting on a hardware security capability they haven't turned on yet. That gap is exactly what attackers are counting on. 👨💻 👇🏼Click below to learn how hardware-backed device identity works! 👇🏼https://hubs.ly/Q04lMNTl0

Be honest, how many SSH keys are just vibing in your infrastructure right now? 🔑 Here's an audit most teams don't run often enough: grep your authorized_keys files, check your CI/CD pipelines, look at what's sitting in your secrets manager. Count the keys that haven't been rotated in over a year. Count the ones that belong to people who left the company. That number is almost always higher than anyone expects. SSH keys have no native expiration. There's no automatic revocation. There's no built-in audit trail that tells you when a key was last used, or from where. Once a key is distributed, you're trusting that it stays secure forever... on every machine it was ever copied to. SSH certificates solve this at the architecture level: 🔑 SSH keys: no expiration, no revocation, no visibility into usage 📜 SSH certificates: short-lived, auto-renewed, tied to identity, auditable by default The common objection is "managing cert issuance sounds painful." It doesn't have to be. Automate it once and expiration stops being a liability, it becomes your best revocation mechanism. A compromised cert that expires in 24 hours is a dramatically different incident than a compromised key with no expiration date and no clear owner. One of these is a security model. The other is a credential management problem waiting to become a breach post-mortem. Learn more here! https://hubs.ly/Q04kBT-f0

Your AI agent has all the access with zero of the accountability. 🫣 In 2026, AI agents aren't just answering questions, they're executing workflows, calling APIs, reading sensitive data, and making autonomous decisions. The attack surface is real and it's growing fast. Here's the problem: most AI agents and MCP servers authenticate with static API keys or long-lived tokens. Credentials that don't expire, can be copied, and grant the same access whether they're being used by your legitimate agent or someone who found the key in a .env file on GitHub. This isn't a new problem. It's the service account problem from 2015 back for its sequel. We know exactly how that story ends. 📖 Short-lived, hardware-backed certificates give every agent a cryptographic identity: who it is, what device or runtime it's running on, what it's authorized to do, and when that authorization expires. No implicit trust. No "we'll rotate the key eventually." No credentials that outlive the incident they were meant to prevent. If your AI pipeline can't answer "which agent is this, running on what, authorized by whom, for how long?" You don't have a security model. You have a trust assumption dressed up as infrastructure. We're breaking down exactly how to secure your AI stack before it becomes a headlines situation at our MFA for AI webinar on June 30th. Register below! 👇 https://hubs.ly/Q04kvPtY0

Fantastic session, led by two of our absolute best. Definitely worth the watch/listen. #Cybersecurity #AI #DeviceIdentity #AutonomousSystems #IdentitySecurity #DefenseTech

"We Know This Device" and "We Can Prove This Device" Are Not the Same Thing 🙅♀️ There's a difference between "we know this device" and "we can prove this device" and most organizations have been quietly conflating the two for years. 🔍 Here's what "device identity" looks like at most companies in 2026: - Device enrolled in MDM ✅ - Has a certificate (maybe) ✅ - That certificate may or may not be hardware-bound 👀 - And it could have been exported to a completely different machine 🫠 Enrollment tells you a device was configured at some point. It doesn't tell you which physical machine is making a request right now. This distinction matters because modern attackers aren't blasting through firewalls. They're presenting valid credentials from the wrong place, on the wrong device, often at 2am. Real device identity requires hardware attestation: a cryptographic guarantee that a private key never left a specific TPM or Secure Enclave. You can't export it. You can't copy it. You can't replay it from a different machine. The question isn't "is this device in our system?" It's "can we prove this is the device we think it is, right now, cryptographically, every single time?" Those are very different security postures. One of them actually stops modern day attacks. Link: https://hubs.ly/Q04kxkBv0

☀️ The sun is out. The grill is fired up. Vacation plans are on the calendar. That can only mean one thing: it's officially summer listening season! 🎧🍔 We asked a few Smallsteppers what cybersecurity podcasts are making it into their headphones this summer, and these were some of the favorites: Security Now Security. Cryptography. Whatever. Root Causes Podcast Cloud Security Podcast Mac Admins Podcast CISO Series Whether you're road-tripping to the beach, mowing the lawn, sitting by the pool, or watching your kids on the playground, these podcasts are packed with great conversations on security, identity, cryptography, cloud security, and the future of technology. What cybersecurity podcasts are on your summer playlist? Drop your favorites below 👇 #Cybersecurity #InfoSec #PodcastRecommendations #SecurityPodcast #SummerListening

With Autonomous systems, authentication happens once, but execution happens forever. ❌ The full recording of our webinar, Securing Autonomous Defense Systems with Hardware-Attested Device Identity, is now available on demand! In this session, J. Hunter Hawke and Josh Drake explored how autonomous systems, AI workloads, and machine-driven environments are reshaping modern security assumptions and why traditional identity models built around humans, passwords, and one-time authentication events are no longer enough. We covered: - Why the human-in-the-loop assumption no longer holds - The growing gap between authentication and execution - Why hardware-attested device identity becomes foundational in autonomous environments - How TPM and Secure Enclave-backed attestation strengthen trust enforcement - What autonomous defense and manufacturing systems mean for the future of security As autonomous platforms continue operating long after the original authentication event, organizations need a way to verify what is actually executing actions, not just who initiated them. 🎥 Watch the webinar on demand: https://hubs.ly/Q04h_T4z0 #Cybersecurity #AI #DeviceIdentity #AutonomousSystems #IdentitySecurity #DefenseTech

For the second year in a row, Smallstep is proud to sponsor Black Hat USA! 😎 Join us August 3rd–6th at Mandalay Bay in Las Vegas and stop by Booth 5111 near the AI Zone to see what’s next for device identity, phishing-resistant access, and securing AI-driven environments without relying on blind trust.✅ We’ll have interactive touchscreen demos at the booth so you can experience firsthand how hardware-attested device identity changes the way organizations secure access across devices, workloads, and modern infrastructure. We’re also hosting an exclusive MFA for AI theatre session at the booth, where we’ll break down why traditional identity models fall short for AI agents, autonomous systems, and non-human identities and what security teams need to do next. As AI agents and autonomous systems continue to reshape enterprise security, one thing is becoming clear: identity cannot stop at the user. Trust has to start at the device. See you in Vegas! 🎰 #BlackHatUSA #Cybersecurity #AI #DeviceIdentity #IdentitySecurity #MFA

This Memorial Day, we honor those who gave their lives in service to our country and reflect on the sacrifices made to protect the freedoms we often take for granted. Thank you to the service members, veterans, and military families whose courage and commitment continue to inspire us. Wishing everyone a safe and meaningful Memorial Day weekend. 🇺🇸 #MemorialDay #HonorAndRemember