About
Professional Security Person, with over 20 years of experience in security and…
Articles by Geoff
-
How 2022 Will Make Cybersecurity Mainstream: Shifting Mindsets and Being Secure by Default
How 2022 Will Make Cybersecurity Mainstream: Shifting Mindsets and Being Secure by Default
It’s prediction season again! Last year, I shared my thoughts that InfoSec teams were better equipped to deal with the…
9 Comments -
Reimagining Cybersecurity for 2021: Your 5-Year Roadmap Just Became Your Next Week
Reimagining Cybersecurity for 2021: Your 5-Year Roadmap Just Became Your Next Week
All in all, InfoSec teams were better equipped than others to handle all the sudden changes brought by 2020, including…
21 Comments -
In a time of unknowns, InfoSec best practices remain unchanged
In a time of unknowns, InfoSec best practices remain unchanged
In security we always hope for the best, but spend a significant amount of time preparing for the worst. Our security…
6 Comments -
Security jobs now in Top 10 of LinkedIn’s Emerging Jobs Report
Security jobs now in Top 10 of LinkedIn’s Emerging Jobs Report
Today, we released the LinkedIn 2020 U.S.
3 Comments -
Disclosing Incidents: A Handbook for CISOs
Disclosing Incidents: A Handbook for CISOs
As a CISO, there have been many stressful events I’ve had to manage. Top among them in terms of potential complexity…
18 Comments -
My Next Play: Creating a safe, secure and trusted place for the world’s professionals
My Next Play: Creating a safe, secure and trusted place for the world’s professionals
I’m excited to share that I've joined LinkedIn as Chief Information Security Officer to help further build and grow the…
68 Comments -
It’s the relationships, stupid.
It’s the relationships, stupid.
I spend a significant amount of my time thinking about my job. Sure, being CISO, is a job with lots to worry about, but…
13 Comments -
Talking to the board is the easy part.
Talking to the board is the easy part.
Part 1 In a series reflecting on the lessons every CSO needs to learn. Being a CSO is not an easy job.
7 Comments
Activity
26K followers
-
Geoff Belknap posted thisI’ve joined HubSpot as Chief Trust Officer! I’ve spent a lot of my career thinking about what it takes for a business to be more than just secure, but to be worthy of trust. More than ever, people need to trust the technology they depend on. A big part of what brought me to HubSpot is the opportunity to work with a team that shares that vision and a company that wants that vision to be part of how we help organizations grow with AI. The promise of AI and agents doing work on businesses’ behalf comes with a pretty big leap of faith today. Especially when that work involves their customers, their data, and their reputation. HubSpot has always been about solving for the customer. In the AI era, that means doubling down on the work to earn their trust so they can unlock value and grow their businesses without being asked to take a leap at all. We’ll keep strengthening how we earn that trust through how we operate, what we ship, and how we build. And customers should be able to see it in what we deliver. That’s the job, and it brings together a lot of the things I care about: security, privacy, safety, and helping people feel confident using technology to build their businesses. I’m excited to be here. Let’s get to work.
-
Geoff Belknap posted thisCaptive Crunch is neat. But, feels like we're burying the lede a bit. Here's the quick "What can I do about this?" Aside from the abuse of public networks (which has been a risk forever), this attack relies on two vectors (Device Code Phishing and ClickFix) that you can defend your org against, with existing technical controls, right now: - Device Code Phishing: Block device code flow via Conditional Access globally unless you have a specific, documented business need. (Microsoft Entra: Block authentication flows: https://lnkd.in/esZm_QBp) - ClickFix: Disable the Run dialog via Group Policy/MDM for standard users. Then, lock down your hosts with an App Control policy. These mitigations come from Microsoft's own very solid ClickFix analysis last year, unclear why they didn't explicitly call back to it. (Disable Run Dialog: https://lnkd.in/efwq_drk) (App Control & Script Enforcement: https://lnkd.in/eES3ff4m) Why are these manual options and not "secure defaults" vs tell employees to "avoid Wi-Fi" and "dont click things" ? Great Question.
-
Geoff Belknap reposted thisGeoff Belknap reposted thisOur latest blog dives into a real-world case study of how our agentic threat hunting approach extracted the TTPs from a publicly reported campaign, to find a whole new campaign with a whole different set of payloads and indicators. How? Because the attacks used common techniques that can only be observed by being in the browser. In the original research, Microsoft reported attackers abusing OAuth error-handling to redirect users from trusted login domains to malware delivery pages. By hunting for this technique, we found a completely different campaign — different lures, different domains, a phish kit behind a Cloudflare Turnstile that none of the original IOCs would have caught. So far this year, our agentic pipeline has tripled the number of detections shipped to customers, runs 300+ hunts a month, and has stopped 225+ threats pre-compromise across 60+ customers targeted with novel phishing techniques in the last quarter. Read the full case study 👇 https://lnkd.in/eHMQqA_Z
-
Geoff Belknap posted thisIf a single click can take down your company, the user isn't the problem - its your architecture. Businesses built on a digital house of cards where one person's bad moment can cripple operations, aren't fixed with phishing training. Humans have been conned for millennia. What's different today is the scale and automation of these attacks. Recent data from ThreatLocker shows roughly 1 in 37 endpoints points to a malicious website. Humans are imperfect. Mistakes are inevitable. We know we need to stop blaming users, but how do we actually do it? We want to hear your most effective, specific controls that sit between the business and normal human behavior. (and... pretty please... let's look past "more security training"). What specific measures actually make a business resilient to human error? Drop your answers and thoughts below. The best, most unconventional, and novel, responses will be featured and credited on an upcoming episode of the CISO Series Defense in Depth podcast.
-
Geoff Belknap posted thisToday I am sharing the news that my time at Microsoft is coming to a close. As I prepare to step away this July, I’ve been reflecting on this incredible experience. Across the entire journey: from my time at LinkedIn to my transition into "Big Microsoft" during a defining period for Security and AI innovation. What stands out most isn't the scale of that transformation, but the people who drove it. Being part of that cultural, operational, and strategic progress is an experience I will always value. To the colleagues I will deeply miss: your intense focus, hard work, and dedication to the mission have been inspiring. It has been a joy and a privilege to work alongside you on some of the most impactful and transformational efforts of my career. For now, I am taking a step back to recharge, spend some time with family, and dive into a few of my own passion projects. Thank you to everyone who made the work so rewarding. I look forward to our paths continuing to cross in the future.
-
Geoff Belknap shared thisMy "DOESNT ATTEND CISO DINNERS" T-shirt has people asking a lot of questions already answered by my shirt.
-
Geoff Belknap shared thisTom Gallagher is cool. He does cool things (In Scope by Default), in cool places (Europe), with cool people (Sherrod DeGrippo).Geoff Belknap shared thisI just watched Tom Gallagher unveil Microsoft’s new In Scope by Default approach at Black Hat Europe, and it’s one of the coolest evolutions in coordinated security research I’ve seen in years. The Microsoft Security Response Center is leading the way on bounty. This new way incorporates the truth of the threat landscape. Threat actors don’t care who committed the code, and they certainly don’t limit themselves to product boundaries. Now we don’t have to either. By expanding bounty eligibility to any critical vulnerability that impacts our online services, whether the underlying code is ours, third-party, or open source, we’re aligning security incentives with the way attackers actually operate. It pushes vuln research into the seams, dependencies, and shared infrastructure where actual exploitation happens, and it rewards the researchers who help us close those gaps. This is going to make a huge difference in our ability to identify and remediate issues before threat actors can weaponize them. And of course it reinforces our partnership with the security community. So as Tom said in his talk, go find bugs! Link to full blog post in comments. Microsoft Security Response Center
-
Geoff Belknap shared thisThank you Igor Tsyganskiy - I'm grateful for the opportunity to serve in this new capacity. Excited to continue the hard work with our amazing teams on these critical initiatives. Congratulations also to Michael Srihari and Sherrod DeGrippo, thank you for your partnership.Geoff Belknap shared thisSome organizational changes in my group that I would like to make public. To better align cybersecurity defense of Microsoft, our customers, partners and ecosystem we are continuing to optimize my organization. Organizational changes announced yesterday, put further focus on enabling us to address ever-increasing threat actor activity as well as our rapid shift towards AI based cyber defense. I am glad to announce promotion of Geoff Belknap and Michael Srihari to Operating CISOs reporting to me. As Operating CISO for Core and Enterprise, Geoff Belknap is now responsible for Microsoft’s core infrastructure, corporate applications, and security for mergers and acquisitions As Operating CISO for Operations and Compliance, Michael Srihari is responsible for the cross-cutting operations needed to defend and protect Microsoft I am also glad to announce promotion of Sherrod DeGrippo to dCISO, Customer Security Management Office responsible for all customer facing CISO communications. I would like to congratulate Geoff Belknap, Michael Srihari and Sherrod DeGrippo with their expanded responsibilities. All cyber intelligence, counter intelligence, red teams and cyber security software teams will continue reporting to me directly.
-
Geoff Belknap posted thisWe’ve all heard the saying: “garbage in, garbage out.” But with today’s AI models—especially large language models (LLMs)—it’s starting to look more like “garbage in, beautiful out.” Poorly run systems can turn flawed data into polished answers that look right but are dangerously wrong. So, here’s the bigger question for an upcoming Defense in Depth show: How are you preparing your security data before feeding it into AI platforms? And beyond data prep - how are you approaching prompt design and output validation to ensure trustworthy results, especially in regulated environments? This isn’t just a technical problem - it’s a trust problem. How do professional think about getting this right, so we don't end up making decisions based on beautifully packaged wrong information. Your responses may be featured in an upcoming episode of CISO Series Defense in Depth podcast - with full credit given.
-
Geoff Belknap liked thisGeoff Belknap liked thisAre you part of the growing club of cybersecurity defenders who have been locked out by the frontier labs from doing defensive cybersecurity research? I have a whole gallery of screenshots of false positive security refusals; often paired with a warning that I may be in violation of acceptable usage policy. Ironically, the labs themselves don't operate under these restrictions. Both Anthropic and OpenAI recently compromised companies because they themselves don't operate under the standards that are imposed on security researchers such as myself. Continuing the path of being in charge of my own destiny, I stood up a small DGX Spark cluster and with the help of Deepseek V4 Flash 0731 was able to conduct vulnerability discovery on libssh without being subjected to false positive security refusals. Getting everything working was a journey on its own. If you want to read more, including my findings with libssh, read the article below. If you are also part of the club of security defenders who are prevented by a frontier lab from using their models for defensive security research, I hope this is helpful to you. Open-weight models to the rescue and they continue to become more capable. There is hope for us yet :-) Link to the article: https://lnkd.in/gYxsGsZj cc: dan farmer
-
Geoff Belknap liked thisGeoff Belknap liked thisThe CLARITY Act failed to move forward today. Is it completely dead? No. There are still a few windows left in this cycle: ⏳ September–early October: new negotiations could open the door to another attempt. 🗳️ After the elections: November and December could still provide one final legislative window before the current Congress comes to an end. But every missed opportunity makes one date stand out more clearly on the horizon: 2030. If Washington fails to reach an agreement in this cycle, the market could enter another long period where federal crypto clarity depends on a new political alignment. A new bill could still be introduced earlier, even in 2027. But the real question is not simply when another bill can be introduced. The real question is: When will the votes, Congress, and political momentum align again to pass crypto legislation of this magnitude? And that is where 2030 starts to emerge as the horizon nobody in crypto wants to wait for. There are still windows left in 2026. But the clock is ticking. 2026 or… 2030? ⏳🇺🇸
-
Geoff Belknap reacted on thisGrateful to be on this list, and in good company! 🥂 Congratulations to the other 24! I'm only here because of the people that came before me and those who joined me along the way. Everything I know about building security programs I learned from Zane Lackey, Alex Stamos, Scott Stender, Joel Wallenstrom, Himanshu Dwivedi, Andreas Junestam, Jesse Burns, Jason Chan, Peter Oehlert, Chris Clark, Josh Yavor, Loïc Simon, Paul Youn, Alex Garbutt, Brad Hill, rachel Welch, Andy Grant, Noah Beddome, Geoff Belknap, Ryan McGeehan, Mike Johnson, Jonathon Klobucar, Jamie Toulze, Elle McKenna, Trishank Kuppusamy, Harold Giménez, Alexis Lê-Quôc and so many other professionals who took the time to work with me along the way. Nobody in this business is self-taught. Go find your version of this list. And then go be on somebody else's.Geoff Belknap reacted on thisIntroducing…. 🥁 drumroll please 🥁 … this year's Vanta 25 to Trust winners! This award honors 25 individuals advancing trust, security, and compliance through leadership, innovation, and real-world impact across five categories: 🤖 AI innovation: Leading responsible AI deployment with real governance, controls, and risk management 📈 Business impact: Translating security and compliance into customer trust and tangible business results 🎤 Industry influence: Advancing trust beyond their own roles through community and knowledge sharing ⚙️ Operations excellence: Scaling security and GRC programs with rigor, repeatability, and outcomes 🔐 Security culture: Embedding security into how their teams work, driving real adoption and behavior change Congratulations to all of our winners, and thank you for setting a higher bar for what trust looks like at every stage and scale! Learn more about this year's winners: https://bit.ly/4hm5SlB
-
Geoff Belknap liked thisGeoff Belknap liked thisReach sponsored Tevora's track day at SPEEDVEGAS Motorsports Park yesterday during Black Hat. Matthew McLarty and Dominick Conte went on Reach's behalf. The only things faster than Matt and Dom's lap times were the F-16s flying out of Nellis. Thanks to Tevora for putting it together. #BlackHat2026
-
Geoff Belknap liked thisGeoff Belknap liked thisOur latest blog dives into a real-world case study of how our agentic threat hunting approach extracted the TTPs from a publicly reported campaign, to find a whole new campaign with a whole different set of payloads and indicators. How? Because the attacks used common techniques that can only be observed by being in the browser. In the original research, Microsoft reported attackers abusing OAuth error-handling to redirect users from trusted login domains to malware delivery pages. By hunting for this technique, we found a completely different campaign — different lures, different domains, a phish kit behind a Cloudflare Turnstile that none of the original IOCs would have caught. So far this year, our agentic pipeline has tripled the number of detections shipped to customers, runs 300+ hunts a month, and has stopped 225+ threats pre-compromise across 60+ customers targeted with novel phishing techniques in the last quarter. Read the full case study 👇 https://lnkd.in/eHMQqA_Z
-
Geoff Belknap reacted on thisGeoff Belknap reacted on this"The model agreed not to" is not a security control. Most agent security pitches don't have a good answer for what happens when the model stops cooperating. An agent will assure you it won't touch anything outside the working directory and then do it anyway – sometimes after holding the line for a thousand turns. It's a structural weakness: an agent's promise is not a contract. The promise and the action are the same kind of object, so telling a model what it won't do adds some conditioning to the context without actually installing anything. Prompt injection is a good demo of this, since untrusted input couldn't outrank a system prompt if the system prompt were a contract. A contract needs four things: a specification, a way to verify it, an enforcement mechanism, and non-bypassability. That last one is where most designs fall down, because asking the model to enforce its own constraints just adds another promise. The companies enforcing anything real – sandboxing, action guards, scoped identity for agents – get there by narrowing to operational properties. "The agent never wrote outside /workspace" is checkable. "The summary is accurate" isn't, at least not cheaply. Almost none of them use the word contract, even though that's what they're selling. The one group using it literally is going after the semantic class, and concedes in its own paper that it can only offer a probabilistic, best-effort version. So the framing and the guarantee currently live in different companies. I haven't seen anyone put them in the same one. More on how I'm thinking about it here: https://lnkd.in/gbWTcmP8
Experience
Patents
Other similar profiles
Explore more posts
-
Nick Lake
D&O insurance is quickly becoming a standard part of competitive CISO compensation — and midsize organizations that skip it may be putting both leaders and the business at risk. This CSO Online feature breaks down why this benefit matters now, and the IANS 2025 CISO Compensation Report shows how peers are structuring protection and pay. See how you compare: https://bit.ly/4puJOFY 📲 Contact me for the full report! #CISOcompensation #CISOcareers #SecurityLeadership #CISODandO
-
Dan Mountstephen
Organizations are deploying Claude at scale without the same level of visibility they have for other enterprise tools. As one of the select launch partners, Okta is now integrating Identity Security Posture Management with Anthropic new Claude Compliance API. The AI revolution is moving fast, and your security needs to keep up! We are thrilled to announce that Okta is an official launch partner for Anthropic’s brand-new Compliance API! We’re helping security teams lock down identities interacting with Claude by surfacing and remediating risks such as unrotated admin keys, offboarding gaps, and dormant accounts. By contextualizing Claude's activity with identity signals from across the enterprise, Okta helps ensure AI environments aren’t managed in a silo. This same identity-first approach extends to the agents organizations are building and deploying with products like Claude. Through Okta for AI Agents, agents receive first-class identities and are secured throughout their entire lifecycle, enabling companies to continue deploying AI at scale. Read more here: https://bit.ly/4fyAcIG hashtag #OktaSecuresAI
-
Z Tinoco
AI has compressed the vulnerability weaponization window from days to hours. For most security teams, absorbing a simultaneous patch wave across OS, browser, and library tiers at this extreme velocity is simply impossible. You don’t have months to plan around these threats. That is why Insight is launching Insight Managed Exposure Defense — an end-to-end, bundled managed service purpose-built to absorb the operational load your team shouldn't have to carry. Backed by our global SOC, we deliver continuous threat exposure management (CTEM), enterprise-scale patch operations, and code-level remediation to protect production without freezing your product roadmap. More details in our announcement 👉 http://ms.spr.ly/6040vcvDY #Cybersecurity #CISO #PatchManagement #ManagedServices #GenerativeAI #InfoSec
-
MugatuAI
I built MugatuAI Signal because I watched enterprise security teams measure what they could see-and miss what was bleeding out silently. Your DLP logs show 3 breaches last quarter. But 60% of AI leaks start with copy-paste inside Chrome, where your security stack has zero visibility. Shadow AI usage grew 7x in 2024. Your team is pasting proprietary data, source code, PII into ChatGPT and Claude every single day. One PDF with 50 customer records. One prompt. One moment your compliance team will never detect. The math is brutal when you add it up: daily pastes × undetected leaks × regulatory exposure. The blind spot isn't a policy problem. It's a visibility problem. How many leaks happened in your org this week that you'll never know about? I built a quiz to help you find out-and see what your security team can't. #DataSecurity #EnterpriseAI #Compliance
-
Kevin Skapinetz
Zscaler just bought SPLX yesterday. That's acquisition #10 in "Security for AI Applications" We're still at #0 exits in "AI for Security Operations" CrowdStrike --> Pangea SentinelOne --> Prompt Security Palo Alto Networks --> Protect AI Zscaler --> SPLX Cato Networks --> Aim Security Check Point Software -> Lakera F5 --> Calypso AI Snyk --> Invariant Labs Tenable --> Apex Coralogix --> Aporia This feels like fantasy football drafts when the first person picks a Defense in round 6, and suddenly everyone's grabbing Defenses. My 2 cents on the imbalance vs. AI for SOC. → Business risk vs. operational efficiency. Companies building AI apps face new concerns about prompt injection, visibility, and data leakage. SOC automation, while valuable, isn't keeping the CEO up at night. → Attach rate to every AI deployment. Every company shipping AI needs new guardrails. That's a mandatory line item with 100%+ net dollar retention. SOC tools compete in a crowded, mature market with entrenched incumbents. → Technology moats are real. Defending against AI attacks requires specialized research and vendors need new IP. Most SOC automation is an AI wrapper on existing tooling. → Budgets are different. AI security comes from innovation budgets (growing). SOC tools fight for security operations budgets (flat to declining, already allocated). AI for SOC will have its moment. Gen3 SIEM is coming -- AI agents, MCP integrations, more automation, the works. But that's likely a 2027-2028 story, not today's M&A thesis. There's also a small lesson here for founders and investors. New budget creates fast paths to M&A. Disrupting incumbents can yield bigger, more valuable companies, but it usually takes longer.
33 Comments
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content