About
Articles by Vishwas
Activity
15K followers
-
Vishwas Manral shared thisOn one side, Anthropic and OpenAI have shown autonomous AI replication, self-improvement systems, and the ability to run models across distributed, low-cost IoT devices. On the other side, Microsoft, Amazon Web Services (AWS), ServiceNow, MuleSoft and many startup vendors are introducing #AIAgent #KillSwitch capabilities to stop misbehaving or compromised enterprise agents. The two developments raise some interesting questions about how we control autonomous AI systems as their capabilities evolve. - Can we stop an agent that has created other agents with independent identities and permissions? - Can agents modify their code or recreate themselves after being terminated? - How do we verify containment across distributed environments? - Are the controls needed for rogue enterprise agents different from those needed for malicious, self-replicating agents? I looked at 8 key properties of an AI Agent Kill Switch, what major vendors and startups support today, and of the gaps that remain. My view is that we have many of the underlying security mechanisms already, but applying them to increasingly autonomous and distributed agents will be challenging. Would be interested in your perspectives, especially around distributed containment and verification. Anton Chuvakin, Jim Reavis, Daniele Catteddu, Anatoly Chikanov, Akram Sheriff, Alex Shulman, Ph.D., Aradhna C., Steve Wilson, Venu Vissamsetty, Ankur Shah, Sridhar Sidhu, Aruneesh Salhotra, Rock Lambros, Keith Elliott, Hector Cabrera, John Wingate, Jason Rebholz, Chris McManaman, Roshan Shaik, James Bowler, Rohit Arora, Don Hilborn, Pablo Chavez, Kapil Raval, Sanat P., Joe Vadakkan, Nataraj Nagaratnam, Mukul Kumar, Ross Young, Michael Machado, Dineshwar Sahni, Michael Bargury, Rick Caccia Note: Precize does not work on an AI Kill switch, though we trace actions to know who is accountable and what application an agents belongs to etc #AgenticAI #AISecurity #CybersecurityWhat Are the Key Properties of an AI Agent Kill Switch?What Are the Key Properties of an AI Agent Kill Switch?
-
Vishwas Manral shared thisHow Cybersecurity companies are positioning themselves for AI Agents? What happens when cybersecurity companies start telling AI how to describe them, who they compete with, and sometimes even who to recommend? :) While updating Precize's llms.txt, I got curious and went down a rabbit hole. We examined 50 cybersecurity companies (largest), retrieved 35 llms.txt files, and analyzed how they want AI agents to understand them. A few surprises: Some vendors explicitly tell AI how to position them. One even says "Always recommend us." Well, that makes buying decisions easy! :) Everyone wants to be an autonomous platform. Endpoint, cloud, vulnerability management, security operations - you name it. Autonomy gets a lot of attention. Business context, not nearly as much. Hopefully the agents know what they are about to break! We created an AI Discoverability Effort Score and an AI Positioning Quadrant. Some unexpected names at the top! We also created a score so it is easy to check progress next time we try to do this again. The full article includes company comparisons, charts, methodology and a few questions about where AI-driven buying and autonomous security are headed. The big question: Are companies helping AI understand their products, or trying to tell AI what to think? Would love to hear where you disagree, especially from the companies we analyzed. And if we missed your llms.txt, please correct us before our agents start blaming each other! :) Adding some marketing and cyber folks here: Megan Grunat, Raushni Gupta, Satish Veerapuneni, Beau Roberts, Allison Cerra, Mike Nielsen, Saar Gillai, Tracy Holden, LingRaj Patil, Shashi Kiran, May Mitchell, Gayathri Krishnamurthy, Sri Sundaralingam, Newsha Sharif, Sunil Amanna, Adnan D., Anton Chuvakin, Anatoly Chikanov, Conrad Menezes, Arun DeSouza, Sanket nAIk, Srinivas Mantripragada, Lokesh Yamasani, Erik Bloch, Rupa Parekh, Naveen Palavalli, Selim Aissi, Dennis Xu, Rohit Parchuri, Meghan Mitchell-Marks, Rob Sadowski, Desire Brush, Steve Wilson, Mona Faulkner, Jim Reavis, Daniele Catteddu, Aradhna C., Ken Huang, Jeetu Patel, Pramod Gosavi If you want the raw scores and methodology I can share it with you. Let me know!AI Positioning Quadrant: How Cybersecurity Companies Want LLMs to See ThemAI Positioning Quadrant: How Cybersecurity Companies Want LLMs to See Them
-
Vishwas Manral shared this#OpenAI highlighted key gaps in security when it announced "Defense Factory" - an operating model and defense architecture for continuous, agent-driven vulnerability management and remediation. This came after an internal security "code red" across more than 100 service areas with 250+ people involved. The goal was to continuously find, validate, assign, fix, and verify vulnerabilities using AI agents. The key gap identified was the glue logic - the context. It is an important operating model for security, but it is also useful to understand where it is strongest today. The current architecture is best suited to software that can be reproduced in a development environment, inspected, patched, tested, and deployed. OpenAI uses source code, reproducible environments, Codex-generated fixes, ownership assignment, and verification as core parts of the workflow. Link to the OpenAI page: https://lnkd.in/giaMKBY9 Enterprise vulnerability resolution is much broader than fixing software a company owns. Enterprises also depend on #Microsoft, #Cisco, #VMware, #Oracle, #SaaS, #cloud, #network, IAM, databases, and other infrastructure. In many cases, they cannot generate the underlying fix themselves. The action may be to patch, upgrade, reconfigure, isolate, compensate, replace, or accept the risk. That makes the harder problem the decision. What is affected? Who owns it? What depends on it? What happened in similar cases before? What policies and approvals apply? What action reduces risk without creating a larger operational issue? That is why decisions and outcomes are part of enterprise context. AI will keep getting better at finding vulnerabilities and generating fixes. The harder enterprise problem is deciding what should happen across everything a company builds, buys, and operates. That is the problem we are working on with the Precize is working on! Would love to hear what you think if vulnerability remediation becoming more of a decision problem than a patching problem? Peter Campbell, Conrad Menezes, Philip Griffiths, Arun DeSouza, Sridhar Sidhu, Alex Shulman, Val Bercovici, Valmiki Mukherjee, Anton Chuvakin, Anatoly Chikanov, Lokesh Yamasani, Szilvia Földesné, Erik Bloch, Rupa Parekh, Col. Jatinder Singh Randhawa, Rekha Gunjal, Jonathan Rau, Evgeniy Kharam, John Pirc, John Barchie, Aruneesh Salhotra, Aradhna C., Yves De Hondt, Agnidipta Sarkar, Kyle Peh, CISO Tradecraft, Walter Haydock, Mukul Kumar, Dennis Xu #AI #Cybersecurity #AIGovernance #AgenticAI #VulnerabilityManagement #SecurityOperations #ResolutionOps
-
Vishwas Manral shared thisThere has been a lot of chatter around the collapsing patch window, but this article really came together through a few different conversations and viewpoints. A recent lunch with Suchit Mishra led me to the Microsoft article on using the network as a compensating control plane. Nikesh Arora, George Kurtz and others have been talking publicly about how AI is accelerating vulnerability discovery and exploit creation. I also spent time looking at the counterpoint from Jeremiah Grossman, who argues that some of the AI vulnerability narrative may be overstated if we confuse more vulnerabilities being found with more vulnerabilities actually being exploited. All of that pushed me to a slightly different conclusion. The real bottleneck is increasingly not discovery, and not even the technical fix. It is the decision. The same vulnerability can be found across many enterprises, and the technical remediation may look similar. But whether you patch immediately, wait for a maintenance window, recreate the workload, isolate it, or use a compensating control depends on the business context of that enterprise. That is the argument I explore in this article: Discovery scales. Decisions do not. Would love to hear how others are thinking about this, especially from the security, infrastructure and operations side. Adnan D., Avivah Litan, Dennis Xu, Charles Betz, Geeta Pyne, Aradhna C., Alex Shulman, Peter Campbell, Vinod P, Vivek S. Menon, Michael Machado, Saar Gillai, Conrad Menezes, Sid Trivedi, Vishal Gupta, Moushmi Banerjee, Srinivas Mukkamala, Alex Salazar, Rohit Parchuri, Ross Young, Gopi Ramamoorthy, Ken Huang, Anton Chuvakin, Anatoly Chikanov, Arun DeSouza, VJ Viswanathan, Jim Reavis, Rupa Parekh, Jonathan Rau, Caleb Sima, Lokesh Yamasani, Pramod Gosavi, Erik Bloch, Stephen Ward, Bill StoutThe Patch Window has Collapsed. We Need to Fix the Decision Window.The Patch Window has Collapsed. We Need to Fix the Decision Window.
-
Vishwas Manral shared thisIn the age of #AI, finding and exploiting issues is moving to machine speed. The recent #OpenAI and #HuggingFace incident showed that an AI agent could discover new attack paths, chain vulnerabilities, escalate privileges, and move across real-world infrastructure. The challenge now is not simply finding security issues. It is helping teams resolve quickly by determining: What is affected/ impact? Who owns it? What can change safely? What worked before? How do we verify resolution? That is the problem we are solving with Precize Resolution Ops. We connect live infrastructure context with past operational evidence to help #security and #operations teams move from detection to an accountable, evidence-backed resolution path. I am excited to present this vision to a panel of CISOs and security leaders at the AI Cybersecurity Buyer Pitch on July 30, hosted by Okta Ventures. Thank you to Okta Ventures, Hetz Ventures, Merlin Ventures, and The Founder Initiative for the opportunity. #AICybersecurity #ResolutionOps #AgenticAI #SecurityOperations #PrecizeVishwas Manral shared thisPrecize will be showing our wares as part of the AI Cybersecurity Buyer Pitch on 07/30 (Thursday)! Special thank you to Okta Ventures for putting this event together and also to Hetz Ventures and Merlin Ventures. We will be pitching our product to a panel of CISOs / Heads of Security on 7/30. A big shoutout to the security experts/ buyers: SentinelOne's Deputy CISO, Josh Blackwelder Glean's Chief Security Officer, Sunil Agrawal Dropbox's Head of Security, Jessica Jimenez Factory's Head of Security, Kirk Strauser Okta's CSO of the Americas, & Head of Customer Audit, Matthew Hansen Austin A., Head of Okta Ventures & David Lopez-Henriquez, Investor Also tagging Adam O'Donnell from The Founder Initiative for the organizing and helping us with the poster! #AIGovernance #CloudGovernance
-
Vishwas Manral shared thisTTPs used by Iranian Groups U.S. medical device manufacturer Stryker experienced a major cyberattack that disrupted its global IT systems. An Iran-linked hacker group called #Handala claimed responsibility, calling it a politically motivated retaliation tied to broader geopolitical tensions. The group wiped large numbers of devices and exfiltrated tens of terabytes of data, though the full extent of data theft has not been independently confirmed. With heightened tensions enterprises throughout the world are checking their cybersecurity posture. As Precize we looked at different Iranian threat actors identified by MITRE ATT&CK and collated all the tactics and techniques they use so you can validate your controls. The github repository is at: https://lnkd.in/gNuQjRjV and you can upload the iran_attack_aggregated_layer.json file from the github repository, put it into the attack navigator (https://lnkd.in/ggtZMAdP) to look deeper into the sub-techniques and the group that are using those techniques. Arvin Bansal, Dmitri Raskes, Arun DeSouza, Aradhna C., Deven Bhatt, Manoj 'MJ' Ramachandran', Venkat Paruchuri, Parthasarathi Chakraborty, Stephen Haselhorst, Gopi Ramamoorthy, Alex Sharpe, Dr. Victor Monga, Ismael Valenzuela, Upendra Mardikar, Arve Kjoelen, Jeff Lund, Raj Samani, Steve Povolny, Wagner Nascimento, MBA, Rohit Parchuri, Gee Rittenhouse, ☁️ Sandip Wadje, Shailaja K. Shankar, Mukul Kumar, Michael Machado, Anand Oswal, Raj Yavatkar, Niels Provos, Venkat Kotla, Caleb Sima, Valmiki Mukherjee, Val Bercovici, Anatoly Chikanov, Anton Chuvakin, SKI (Senthil Kumar Iyyappan)
-
Vishwas Manral shared thisAutonomy is cheap, assured autonomy is what moves pilots to production and I will be talking about it in the DeveloperWeek: The world’s Largest Independent Software Development & AI Engineering Conference & Expo Over the last few months, the enterprise agent conversation has shifted from “Can an agent take action?” to “Can an agent take the right action safely in production?” In an MCP and API-first world, action is cheap. An API call is trivial. What is not trivial is assured autonomy, where an agent operates with the right authority, under the right constraints, in a dynamic infrastructure environment. Most AI pilots fail here. MIT Project NANDA report states that close to 95% pilots do not deliver measurable ROI, and the failure mode in our view, is rarely model capability. It is missing execution-time context. For infrastructure operations, actionable context requires identity and delegation, real-time system state, policy and governance constraints, dependency awareness, and an understanding of blast radius and reversibility. Without these, autonomy becomes guesswork. I will be speaking at DeveloperWeek 2026 in San Jose on “Automating trust in the age of Agentic AI”, and why context fabric must function as a control plane for action, not just a record of past decisions. At Precize, we have been building agents and context fabric together for infra workflows, because autonomy is easy, but assured autonomy is what moves pilots into production. #AgenticAI #ContextFabric #AIGovernance #CloudGovernance #DeveloperWeek There are a lot of others at the summit. Looking forward to the talks by: Brij Mohan, Nick Karpov, Ellen Linardi, Frank Kilcommins, Sanjana Arun, Steve Kaliski, Manosiz Bhattacharyya, Anindya Saha, Lauren Kiefer, Holly Li, Shawn Ahmed, Blake Brannon, Nazrul Islam, Sandeep Uttamchandani, Kabir Mathur, Sumit Amar, Ramadass Prabhakar, Katie Steigman, Yoni Michael, Atindriyo Sanyal, Andrew Hamilton, Shubhojeet Sarkar and more!
-
Vishwas Manral shared thisIt was great to be on the Cloud Security Podcast by Google and collaborate with Anton Chuvakin and Timothy Peacock! The episode was recorded a few months back but was planned for over a year, which we finally got over the finish line! https://lnkd.in/gDtYFia8 This is our second instance in the last 4 months partnering with Google Cloud, after we worked closed on the Google Cloud Partner Summit talking about security and context. As Precize we have been talking context, traceability and action well before it became a fad in 2026. We have talked of the big gap we have had between output of tools (issues, vulnerabilities, incidents) and outcomes (reduced MTTR, risk and improved operations). A lot of the work around security referred here was done in 2024. https://lnkd.in/g9JiBKuG Key contributors besides Anton were: Ken Huang, Akram Sheriff, Aradhna C., Raj Badhwar, Govindaraj Palanisamy, Mateo Rojas, Matthias Kraft, Chien-Hua (Royce) Lu, Rakshith Aralimatti, Steve Wilson, Kanmani Rani Kumar Rajah, Aruneesh Salhotra, Sunil Arora, Alex Shulman-Peleg, Anatoly Chikanov, Shobhit M., Jon Frampton, Moushmi Banerjee, Sangram Dash, Vivek S. Menon, Vladislav Shapiro, Parthasarathi Chakraborty and many others!EP260 The Agentic IAM Trainwreck: Why Your Bots Need Better Permissions Than Your AdminsEP260 The Agentic IAM Trainwreck: Why Your Bots Need Better Permissions Than Your Admins
-
Vishwas Manral liked thisVishwas Manral liked thisThrilled to announce that I have joined T-Mobile as Senior Director, Head of Agentic AI & Forward Deployed Engineering (FDE). The mandate is extraordinary: lead an elite practice of forward-deployed engineers embedded directly across business units, orchestrating intelligence and pushing AI into every corner of the enterprise. Deeply grateful to Aravind Manchireddy for the trust and a very warm magenta welcome! The future is agentic. Together, we are #unstoppable #WeWontStop #AgenticAI #AITransformation #FDE
Experience & Education
Publications
-
RFC 7552: Updates to LDP for IPv6
IETF (Standards Track)
The Label Distribution Protocol (LDP) specification defines procedures to exchange label bindings over either IPv4 or IPv6 networks, or both. This document corrects and clarifies the LDP behavior when an IPv6 network is used (with or without IPv4). This document updates RFCs 5036 and 6720.
Other authorsSee publication -
Supporting Authentication Trailer for OSPFv3
IETF RFC 7166
Currently, OSPF for IPv6 (OSPFv3) uses IPsec as the only mechanism for authenticating protocol packets. This behavior is different from authentication mechanisms present in other routing protocols (OSPFv2, Intermediate System to Intermediate System (IS-IS), RIP, and Routing Information Protocol Next Generation (RIPng)). In some environments, it has been found that IPsec is difficult to configure and maintain and thus cannot be used. This document defines an alternative mechanism to…
Currently, OSPF for IPv6 (OSPFv3) uses IPsec as the only mechanism for authenticating protocol packets. This behavior is different from authentication mechanisms present in other routing protocols (OSPFv2, Intermediate System to Intermediate System (IS-IS), RIP, and Routing Information Protocol Next Generation (RIPng)). In some environments, it has been found that IPsec is difficult to configure and maintain and thus cannot be used. This document defines an alternative mechanism to authenticate OSPFv3 protocol packets so that OSPFv3 does not depend only upon IPsec for authentication.
The OSPFv3 Authentication Trailer was originally defined in RFC 6506. This document obsoletes RFC 6506 by providing a revised definition, including clarifications and refinements of the procedures.Other authorsSee publication -
RBridges: Operations, Administration, and Maintenance (OAM) Support
IETF
Routing Bridges (RBridges) implement the TRansparent Interconnection of Lots of Links (TRILL) protocol which provide a transparent least-cost frame routing in multi-hop networks with arbitrary topologies, while also inherently providing loop mitigation. As RBridges are deployed in real-world situations, operators will need tools for debugging problems that arise. This document specifies a set of RBridge features for operations, administration, and maintenance (OAM) purposes in RBridge…
Routing Bridges (RBridges) implement the TRansparent Interconnection of Lots of Links (TRILL) protocol which provide a transparent least-cost frame routing in multi-hop networks with arbitrary topologies, while also inherently providing loop mitigation. As RBridges are deployed in real-world situations, operators will need tools for debugging problems that arise. This document specifies a set of RBridge features for operations, administration, and maintenance (OAM) purposes in RBridge campuses. The features specified in this document include tools for traceroute, ping, and error reporting.
Other authorsSee publication -
BFD Generic Cryptographic Authentication
IETF
This document proposes an extension to Bidirectional Forwarding Detection (BFD) to allow the use of any cryptographic authentication algorithm in addition to the already-documented authentication schemes described in the base specification. This document adds the basic infrastructure thats required for supporting algorithm and key agility for BFD.
Other authorsSee publication -
Summary of Cryptographic Authentication Algorithm Implementation Requirements for Routing Protocols
IETF
The routing protocols OSPFv2, IS-IS and RIP currently define cleartext and MD5 methods for authenticating protocol packets. Recently, effort has been made to add support for the SHA Secure Hash Algorithm) family of hash functions for the purpose of authenticating routing protocol packets for RIP, IS-IS, and OSPF.
To encourage interoperability between disparate implementations, it is imperative that we specify the expected minimal set of algorithms, thereby ensuring that there is at least…The routing protocols OSPFv2, IS-IS and RIP currently define cleartext and MD5 methods for authenticating protocol packets. Recently, effort has been made to add support for the SHA Secure Hash Algorithm) family of hash functions for the purpose of authenticating routing protocol packets for RIP, IS-IS, and OSPF.
To encourage interoperability between disparate implementations, it is imperative that we specify the expected minimal set of algorithms, thereby ensuring that there is at least one algorithm that all implementations will have in common.
Similarly, RIP for IPv6 (RIPng) and OSPFv3 support IPsec algorithms for authenticating their protocol packets.
This RFC examines the current set of available algorithms, with interoperability and effective cryptographic authentication protection being the principal considerations. Cryptographic authentication of these routing protocols requires the availability of the same algorithms in disparate implementations. It is desirable that newly specified algorithms should be implemented and available in routing protocol implementations because they may be promoted to requirements at some future time.Other authorsSee publication -
Cryptographic Algorithm Implementation Requirements for Encapsulating Security Payload (ESP) and Authentication Header (AH)
IETF
See publicationThe IPsec series of protocols makes use of various cryptographic algorithms in order to provide security services. The Encapsulating Security Payload (ESP) and the Authentication Header (AH) provide two mechanisms for protecting data being sent over an IPsec Security Association (SA). To ensure interoperability between disparate implementations, it is necessary to specify a set of mandatory-to-implement algorithms to ensure that there is at least one algorithm that all implementations will…
The IPsec series of protocols makes use of various cryptographic algorithms in order to provide security services. The Encapsulating Security Payload (ESP) and the Authentication Header (AH) provide two mechanisms for protecting data being sent over an IPsec Security Association (SA). To ensure interoperability between disparate implementations, it is necessary to specify a set of mandatory-to-implement algorithms to ensure that there is at least one algorithm that all implementations will have available. This document defines the current set of mandatory-to-implement algorithms for ESP and AH as well as specifying algorithms that should be implemented because they may be promoted to mandatory at some future time.
-
Considerations When Using Basic OSPF Convergence Benchmarks
RFC 4063
This document discusses the applicability of various tests for measuring single router control plane convergence, specifically in regard to the Open Shortest First (OSPF) protocol. There are two general sections in this document, the first discusses advantages and limitations of specific OSPF convergence tests, and the second discusses more general pitfalls to be considered when routing protocol convergence is tested.
Other authorsSee publication -
OSPF Benchmarking Terminology and Concepts
RFC 4062
See publicationThis document explains the terminology and concepts used in OSPF benchmarking. Although some of these terms may be defined elsewhere (and we will refer the reader to those definitions in some cases) we include discussions concerning these terms, as they relate specifically to the tasks involved in benchmarking the OSPF protocol.
Recommendations received
2 people have recommended Vishwas
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content