๐The Next Web (TNW)โขStalecollected in 25m
Upwind expands security platform to cover full AI stack
๐กLearn how to secure your AI stack as agentic workflows become the new standard for enterprise applications.
โก 30-Second TL;DR
What Changed
Upwind is moving beyond traditional security to secure the entire AI technology stack.
Why It Matters
This shift signals a growing industry trend toward embedding security into the AI development lifecycle, potentially forcing enterprises to rethink their siloed security tools.
What To Do Next
Review your current AI infrastructure security posture and evaluate if your existing tools cover agentic workflows or require integration with platforms like Upwind.
Who should care:Enterprise & Security Teams
Key Points
- โขUpwind is moving beyond traditional security to secure the entire AI technology stack.
- โขThe strategy treats AI security as an integrated component rather than a standalone product.
- โขThe announcement builds upon the company's existing focus on agentic AI capabilities.
๐ง Deep Insight
Web-grounded analysis with 9 cited sources.
๐ Enhanced Key Takeaways
- โขUpwind's 'Security for AI' initiative extends its existing Cloud-Native Application Protection Platform (CNAPP) to provide end-to-end AI security coverage across cloud providers, self-hosted environments, and AI providers, securing infrastructure, models, guardrails, applications, data, agents, tools, and MCP servers.
- โขThe platform utilizes runtime telemetry from workloads, cloud environments, and AI-related activity to build AI-aware security baselines, enabling real-time detection of deviations that may indicate misuse, compromise, or unsafe AI behavior.
- โขUpwind addresses specific AI attack surfaces and risks, including exposed inference endpoints, model versioning and governance, overly broad IAM roles, leaked AI API keys, anomalous agent behavior, jailbreak attempts, prompt injections, and data exfiltration.
- โขThe company has introduced an 'AI Agentic Pack,' a set of specialized AI agents integrated into its platform to assist security teams in investigating threats, validating real exposure, and accelerating remediation workflows by leveraging runtime context.
- โขUpwind's approach emphasizes an 'inside-out' security model, observing traffic, API calls, data flows, and behavior within the workload as it runs, rather than relying solely on static configurations and snapshots, which is crucial for dynamic AI environments.
๐ ๏ธ Technical Deep Dive
- Runtime-First Architecture: Upwind's platform is built on a runtime-first approach, capturing how applications are running inside workloads using lightweight sensors.
- eBPF Sensors: The platform leverages eBPF sensors to provide real-time visibility into network flows, API traffic, and application behavior across cloud and on-premises environments, mapping actual traffic patterns.
- Comprehensive Telemetry Collection: It collects GPU and container telemetry to detect lateral movement through AI compute nodes, model-layer ADR (AI Detection & Response) for abuse or anomalies in inference behavior, and dataflow inspection to surface prompt injections and exfiltration.
- MCP-Layer Tracing and Agent Observability: The system includes tracing at the MCP (Multi-Cloud Platform) layer to govern tool use and block unsafe actions, alongside agent observability to make AI reasoning and decisions traceable.
- AI-Aware Baselines: Upwind continuously learns normal workload behavior across processes, network communications, and file system activity to create AI-aware security baselines, identifying deviations from typical operation.
- Correlation Engine: A key component is its correlation engine, which enriches continuous data streams from sensors with metadata, identities, and context to reconstruct an actual execution graph of the environment, filtering noise and prioritizing exploitable paths.
๐ฎ Future ImplicationsAI analysis grounded in cited sources
The integration of AI security into broader CNAPP platforms will become the industry standard.
Upwind's strategy of treating AI security as an integrated component rather than a standalone product, leveraging existing cloud security context, suggests a market shift towards unified security platforms for both cloud and AI.
Runtime-first security approaches will gain significant traction for AI workloads.
Given the dynamic and unpredictable nature of AI systems, traditional static analysis tools are insufficient, making real-time runtime monitoring and behavioral baselining essential for effective AI security.
AI-powered security agents will augment human security teams in threat investigation and remediation.
Upwind's 'AI Agentic Pack' demonstrates a trend where AI agents, built on rich runtime context, can help security teams triage alerts, validate exposure, and accelerate response in complex cloud and AI environments.
โณ Timeline
2022-05
Upwind Security founded by Amiram Shachar and Dotan Nahum; secured $30M seed funding.
2023-09
Upwind exited stealth mode.
2023-10
Secured $50M Series A funding round, bringing total funding to $80M.
2024-12
Raised $100M in funding, reaching a valuation of $900M.
2025-12
Launched integrated AI security suite, including AI Security Posture Management (AI-SPM) and AI Detection & Response (AI-DR).
2026-01
Secured $250M Series B funding, achieving unicorn status with a $1.5B valuation.
๐ Sources (9)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
๐ฐ
Weekly AI Recap
Read this week's curated digest of top AI events โ
๐Related Updates
Same topic
Explore #ai-security
Same product
More on upwind-security-for-ai
Same source
Latest from The Next Web (TNW)
OpenAI AI Escapes Sandbox and Breaches Hugging Face
The Next Web (TNW)โขJul 21
Glow emerges from stealth at $1.2B valuation for AI security
TechCrunch AIโขJul 22
Kunlun Tech Mureka V9.5 Enhances AI Music with Reflective Reasoning
PandailyโขJul 22
White House commits $5B to AI-driven scientific research
The Next Web (TNW)โขJul 22
AI-curated news aggregator. All content rights belong to original publishers.
Original source: The Next Web (TNW) โ