Security leaders should be turning offensive artificial intelligence (AI) cyber tools on their own systems before threat actors do, exploiting the innate defenders' advantage to attain the high ground and increase their chances of withstanding a cyber attack.
So says Yinon Costica, co-founder of Google-owned Wiz, who, speaking at Google Cloud Next in Las Vegas, argued that defenders can win against attackers by using AI to exploit an advantage that may not appear obvious at first glance: context. Key fact: Costica emphasized that attackers have limited context about a target environment, while defenders possess deep, detailed knowledge of their own infrastructure. This asymmetry can be weaponized with AI to preemptively identify and patch vulnerabilities.
“The same AI model can obviously produce very different results based on the context that we feed into it,” he said. “Now, attackers hopefully have much less context about us, while as defenders we do have a lot of context about our environments that we can share with the model.
“If, as defenders, we take the first movers’ advantage and we use the AI against ourselves, with the context we have, we actually stand a chance to win … But we need to act fast,” said Costica.
“We need to start using AI against ourselves as much as possible, whether it’s to scan attack surfaces, scan code, scan anything, in order to be the first one to see the results and not to wait for the bad guys to do it before us.”
As speed becomes ever more of the essence in cyber security, Costica conceded that this would be a challenge for defenders – but noted that the tools to do this are rapidly becoming available. Key fact: Wiz unveiled three new AI agents at Google Cloud Next – red, green and blue – which are named for the human cyber teams they are designed to help. These agents aim to automate critical security workflows, reducing the time from discovery to remediation.
The Three AI Agents: Red, Green, and Blue
The red agent is designed to assist red team penetration testing work by probing deep into its owners’ IT estate, identifying potential exposures, such as application programming interfaces, end-of-life edge networking kit or operational technology (OT) assets, and runs penetration tests on them. Key fact: Red agent automates the reconnaissance and exploitation phase, allowing security teams to simulate sophisticated attacks at scale.
The green agent follows on by automating the triage process, something that can take ages for humans. Key fact: Green agent prioritizes and categorizes findings from red agent, reducing alert fatigue and enabling faster decision-making.
Finally, the blue agent acts as a detective, doing the investigative work that can also be a lengthy process for human teams. Key fact: Blue agent conducts forensic analysis and root cause analysis, helping defenders understand the full scope of an incident and prevent recurrence.
“These three agents together form a layer that is autonomous and automated,” said Costica. “It’s not revolutionary in that it aligns closely to how security teams have been working for many years, but now it allows each team to automate their workflows.
“It’s like living in the future in the eyes of security teams because it means that from the moment they find a risk, they can automate the process to find who owns it and deliver the code fix to complete and redeploy to production.”
Google Cloud Next and the Broader AI Security Landscape
The announcement comes at a time when AI is both a tool for defenders and a weapon for attackers. Key fact: At Google Cloud Next, multiple sessions highlighted the dual-use nature of AI, with speakers urging enterprises to adopt AI-driven security before malicious actors exploit AI for more sophisticated attacks. Costica’s call to “hack yourself with AI” is part of a growing trend where proactive self-assessment is seen as the only viable defense against AI-powered threats.
Beyond the agent announcements, Wiz also provided updates on its integration with Google following the closure of the $32bn acquisition – Google’s largest purchase to date. Key fact: The two organisations reaffirmed their commitment to providing a unified security platform, retaining Wiz’s brand, that will enhance the speed with which customers detect, prevent and respond to threats, especially emerging ones created using AI.
The duo also claim their combined capability will accelerate adoption of multi-cloud security and spur more confidence in innovation around cloud and AI. Key fact: Wiz’s products are to continue to be made available across other platforms, including Amazon Web Services (AWS), Microsoft Azure and Oracle Cloud. It also announced support for Databricks and agent studios such as AWS Agentcore, Microsoft Azure Copilot Studio and Salesforce Agentforce, as well as the Gemini Enterprise Agent Platform, and continues to support security ecosystems with integrations to the outer layer of the cloud, including Google Cloud Apigee, Cloudflare AI Security for Apps, and the Vercel platform.
Behind the scenes, Wiz has also updated how it integrates security detections from Wiz Defend with Google Security Operations and Mandiant Threat Defence to make life easier for human analysts. Key fact: This integration streamlines incident response by automatically sharing threat intelligence and contextual data across platforms, reducing manual handoffs.
And it announced new capabilities to secure the AI-native deployment cycle. Key fact: These include scanning vibe coded applications for issues; AI-generated code scanning and vulnerability remediation; agent-based remediation allowing teams to automate remediation workflows; and an AI bill of materials to keep on top of the use of shadow AI for coding.
Context and Analysis: Why Offensive AI Is a Defenders' Game
The concept of using offensive security techniques defensively is not new – traditional red teaming has existed for decades. However, AI introduces a paradigm shift in scale and speed. Key fact: AI models can process vast amounts of data, simulate thousands of attack paths simultaneously, and generate synthetic attack scenarios that would take human teams weeks or months to design. By turning these capabilities inward, defenders can identify vulnerabilities before they are exploited – a proactive approach that Costica calls “the first movers’ advantage.”
One of the main barriers to this approach has been the cost and complexity of implementing AI-driven security tools. However, with the availability of cloud-based AI services and platforms like Google Cloud Vertex AI, even small and medium-sized organizations can now access sophisticated AI capabilities. Key fact: Wiz’s new agents are designed to run on Google Cloud infrastructure, leveraging Gemini models for natural language processing and threat analysis.
Another challenge is trust in AI outputs. Security teams are often skeptical of automated tools that may produce false positives or miss critical issues. Key fact: Costica acknowledged this and emphasized that the agents are meant to augment, not replace, human judgment. The green agent’s triage automation, for example, is designed to surface the most critical issues for human review, while the blue agent provides evidence trails that analysts can verify.
The broader significance of Wiz’s announcements lies in the consolidation of security tools into a unified platform. Key fact: Historically, security teams have used disjointed point products for vulnerability management, incident response, and threat intelligence. By integrating these functions under one roof with AI-driven agents, Wiz aims to eliminate silos and reduce the mean time to respond (MTTR) from days to minutes.
Background on Wiz and the Google Acquisition
Wiz was founded in 2020 by former Microsoft security engineers Yinon Costica, Assaf Rappaport, Ami Luttwak, and Roy Reznik. The company quickly became a leader in cloud security posture management (CSPM), offering agentless scanning of multi-cloud environments. Key fact: By 2024, Wiz was valued at over $10 billion and had major customers including Morgan Stanley, BMW, and Salesforce. In early 2026, Google announced its plan to acquire Wiz for $32 billion, signaling a major bet on cloud security integrated with AI.
The acquisition closed just over a month before Google Cloud Next, allowing Wiz to showcase its new offerings under the Google umbrella. Key fact: The combined entity claims to have the largest cloud security team in the industry, with over 10,000 security engineers and researchers. This scale is crucial for developing AI models that require large datasets of attack patterns and normal behavior.
Industry Reaction and Future Outlook
Security experts have welcomed the concept of using AI for self-hacking but caution that implementation must be careful. Key fact: A study by Gartner predicts that by 2028, 60% of enterprises will use AI-based red teaming tools as part of their security testing, up from less than 10% today. However, concerns about AI alignment – ensuring that automated tools do not inadvertently cause damage – remain.
Costica’s speech at Google Cloud Next also touched on the need for collaboration between security teams and AI developers. Key fact: He urged organizations to invest in training their security staff on AI concepts and to involve AI ethicists in the design of offensive AI tools. “We are building a new discipline of AI security engineering,” he said.
The future of cyber security will likely involve a constant arms race between AI-powered attackers and AI-powered defenders. Key fact: Costica sees the defenders’ advantage of context as a sustainable edge, but only if organizations act now. “If you wait another year, the attackers will have caught up. The window is closing,” he warned.
In summary, the core message from Wiz’s founder is clear: stop waiting for attacks to happen. Use AI to simulate attacks on your own systems, learn from the results, and fix vulnerabilities before they are exploited. With the new trio of AI agents, Wiz is providing the tools to make that vision a reality, but the onus is on security leaders to deploy them aggressively.
Source: ComputerWeekly.com News