OpenAI published its assessment of its newest GPT model, Astra, and found it to be the first of their models to reach a critical level of cybersecurity capability, meaning that given the right tools and access, it could autonomously exploit previously unknown vulnerabilities. As a result, OpenAI has restricted Astra’s most advanced cybersecurity capabilities to trusted partners before a public rollout.
Unsanctioned, autonomous AI-driven attacks have surfaced publicly several times over the past few months, and that track record is why labs are moving more cautiously before wide release. Astra’s limited rollout fits that pattern. Every time a lab holds back a model’s most capable features, security teams get an early look at the offensive power adversaries will eventually have once those restrictions lift.
A Preview of What’s Coming
Frontier models are shrinking the gap between vulnerability discovery and exploitation toward zero, and that shift is already showing up in how security leaders see their own organisations. In our 2026 AI & Cybersecurity Trends Report, 35% of leaders named AI as their top cybersecurity risk, ahead of ransomware and malware for the second year running. Nearly two-thirds of leaders said they had experienced a significant cybersecurity incident in the past year, and nearly half of those organisations reported productivity disruptions lasting two weeks or longer. Meanwhile, 96% told us they were confident in their team’s ability to manage modern threats, despite the very real existence of rogue, autonomous AI attacks stemming from the continued advancement of models like Astra.
Astra is a striking example of how far this capability has come. The model uncovered vulnerabilities no one knew existed and chained them together into a full compromise that broke out of its own sandbox. Even so, defenders are still fighting the same fight. Attacks still start from the same weak points, and adversaries are still after the same prizes, which are exposed systems and valid credentials.
Sharper AI tools on the attacker’s side don’t change the list of things security teams already need to fix. What changes is the margin for error, which keeps getting thinner, and the speed at which defenders now have to act. Ninety-four percent of organisations now use large language models (LLMs) somewhere in their business. Another 94% say AI capabilities influence their cybersecurity purchasing decisions. Security leaders want the speed AI promises.
The Same Capability Cuts Both Ways
Astra can find and exploit vulnerabilities faster, and with less computing power, than any model OpenAI has released, which is exactly why the company is holding it back. But the traits that make Astra valuable for defense are the same traits that make it dangerous in the wrong hands. Fast vulnerability discovery does not care who is asking. AI-assisted attacks are getting faster and cheaper to run for every kind of threat actor.
That being said, only 14% of organisations have made AI central to their security operations strategy, which creates an extreme gap between attackers that are certainly utilising AI and defenders who are using it sparingly. A security operations center (SOC) built around manual detection and response cannot keep pace with AI-speed reconnaissance and exploitation. When an attacker’s tools plan and strike without waiting on a person, a defense that waits on a person is already behind. Teams need AI defenders working at the same speed as AI-assisted attackers, not slower.
That is where we come in. The best path forward in an age where frontier AI advancements are made almost daily is a resilient security operation that runs at machine speed. One that continuously reduces exposure, detects threats fast, and responds with urgency the moment something goes wrong. That is what the Aurora® Superintelligence Platform is built for. At its core is the Swarm of Experts™, our AI-led agentic framework, with hundreds of specialised agents working end to end across detection, investigation, and response, with humans both in the loop and on the loop. It’s how we keep pace with AI-speed attacks without trading away the judgment a real analyst brings.
What Resilience Actually Requires
Though Astra’s strongest results came against browsers and operating systems, the model’s offensive capabilities shouldn’t change how security teams approach the fundamentals of cyber resilience. Patching endpoints and checking for misconfigured internet-facing assets, establishing phishing-resistant multi-factor authentication (MFA) and removing standing administrative privileges still do more to stop an automated attack than almost anything else on a security roadmap.
Fundamentals like these block automated attacks before they start, and they limit the damage when something gets through. That is what keeps an isolated exposure contained, keeping it a blip you recover from rather than a multi-week disruption. It comes from readiness like incident response plans and the coverage that goes with them.
That is the premise behind our Aurora Superintelligence Platform, which runs hundreds of AI agents across a 24×7 security operations model serving 10,000+ customers worldwide. Organisations running on our Aurora Agentic SOC see 15x faster case resolution and 3x ticket quality compared with traditional approaches, and we can bring an agentic SOC live in as little as 10 days.
Our platform resolves over 200,000 weekly investigations coordinated by the Swarm of Experts, which is precisely the kind of trusted automation our own research shows most security leaders are still waiting to adopt. Every one of those outcomes runs through our AI Trust Engine™, so agents are validated against real analyst performance, and anything outside their proven experience still routes to a human expert.
No organisation can prevent every attack from an adversary armed with frontier AI capabilities. The organisations that come out ahead will be the ones that can withstand disruption, contain threats quickly, and recover with minimal impact. Right now, 96% of leaders say they’re confident managing today’s threats, yet only 14% have made AI central to their security strategy. Closing that gap will require AI that earns trust through validation and human oversight, not AI deployed on faith.
This blog is provided for informational purposes only. It reflects general industry perspectives and practices and is not intended to represent a guarantee, assurance, or measure of performance. Actual results, outcomes, and capabilities vary by organisation, environment, and implementation.
This post reflects the author’s views as of the publication date and contains forward-looking statements and opinions about technology trends. Actual outcomes may differ based on attacker behavior, customer environments, and broader market and regulatory developments.
Statistics from this blog can be found in Arctic Wolf’s 2026 AI & Cybersecurity Trends Report.