Arora advocates virtual patching, a proven technology available from security vendors, as the most obvious interim solution to AI-accelerated vulnerability discovery.

The most obvious solution is virtual patching. This is a time tested technology that is currently available with cyber vendors for endpoints, WAF, cloud, network use cases.

Zhou warns that AI agents make vulnerability discovery cheap and automated, upending the offense-defense balance.

AI has reshaped the rules of cybersecurity. Vulnerabilities used to be hard to find, so attacks were infrequent and offense and defense were roughly balanced. Now agents can find vulnerabilities in bulk, automatically, cheaply and quickly; without agent infrastructure of equal capability, information systems will be vulnerable everywhere.

Original (zh)

AI已重塑网络安全的游戏规则。过去漏洞很难被发现，因此攻击频率较低，攻防能勉强维持平衡。但现在智能体可以批量、自动化、低成本、快速发现漏洞，如果没有具备对等能力的智能体基础设施，信息系统将处处是漏洞、处处可攻击，面临更大威胁。

Zhou advocates 'using models to govern models', with security agents monitoring other agents while humans retain final say.

The smarter agents are, the less controllable and more uncertain they become. We need 'using models to govern models': build another security model and security agent to analyze agent behavior and inference logs and predict loss of control, reporting to humans who make the final decision.

Original (zh)

智能体越聪明，越不可控、越具有不确定性，它会产生自己的“想法”，此时就需要“以模治模”，即搜集智能体的行为日志、推理日志，打造另一个安全大模型和安全智能体对此进行分析、判断，对它可能的失控行为预判。

Zhou Hongyi revealed 360 found permission and execution risks in the OpenClaw architecture and switched to architectures inspired by Codex and Claude Code, stressing agent security.

周鸿祎透露，团队早期曾尝试OpenClaw架构，但在内部使用过程中发现了权限和执行方面的风险，后来调整为借鉴Codex、Claude Code等产品的架构。

George Kurtz argues that AI safety is what unlocks AI's full benefits, and that deploying agents with a kill switch keeps organizations in control.

“AI safety is how you get the full benefit of AI,” said CEO George Kurtz in a press release. “When you deploy agents with a kill switch, you stay in control.”

George Kurtz said every company is asking whether AI agents can be stopped, and that CrowdStrike's AI kill switch now makes the answer yes.

Every company is asking the same question about AI agents: can they be stopped? Now the answer is yes,

Kurtz says CrowdStrike's AI kill switch gives AI agents the off button they lacked and will be a crucial part of running agents safely.

"Everyone is building AI agents, but they never had the off button, and now they do," Kurtz told CNBC's Jim Cramer on "Mad Money." "It's going to be an incredibly important part of making sure that AI agents run and operate safely."

Kurtz sees securing AI agents as a massive market opportunity, combining external breach prevention with controls on agent behavior.

"Enabling agents to work safely is a massive market opportunity for CrowdStrike," Kurtz said. "We have AI capabilities to prevent these breaches from happening on the outside, and we also have the capabilities to make sure that agents are basically doing what they are instructed to do."

Kurtz likens the kill switch to airbags, saying safeguards are needed even when AI is designed to work correctly, and agents breaching guardrails can be killed at runtime.

"If they go outside of those guardrails at runtime, we can kill those agents," he said.

Kurtz says enterprise customers demanded CrowdStrike integration with Anthropic before deploying AI, signaling strong demand for its security.

"Our customers went to Anthropic and said, 'We need CrowdStrike before we can deploy AI,'" Kurtz said. "I think it says volumes that customers would not roll out AI without CrowdStrike."

Kurtz argues that AI safety, enabled by tools like a kill switch for agents, lets enterprises capture AI's full benefit, making security both accelerant and safety net for adoption.

AI safety is how you get the full benefit of AI. When you deploy agents with a kill switch, you stay in control. For the first time, security is simultaneously the accelerant and the safety net for technology adoption.

Kurtz argues that AI safety is what unlocks AI's full benefits, and that deploying agents with a kill switch keeps organizations in control.

AI safety is how you get the full benefit of AI. When you deploy agents with a kill switch, you stay in control,

Kurtz said that a kill switch for AI agents will become mandatory for companies, as CrowdStrike unveiled a tool to detect, monitor and stop agents.

“I think the kill switch will be mandatory for companies,” Kurtz said.

Kurtz argues AI agents are the new apex predator, letting any adversary operate with nation-state capability.

Now there's a new apex predator and that's the agent. I call this the rise of the agent state. Because when Apex capability becomes a prompt, there are no tears at all. Every adversary, every ecime crew, every insider now can operate like a nation state.

Kurtz says attacks now run at inference speed, eliminating breakout time and making zero-days into 'minus one days'.

Attacks now happen at inference speed. Inference speed. And when the attacker has inference speed, there is no breakout time. There's actually no time at all. And unfortunately, we won't be calling these zero days. We'll be calling them minus one days.

Kurtz says slowing AI development won't fix risk since frontier and open-weight models are already in adversaries' hands.

The challenge we have now is the genie is out of the bottle. Slowing down future AI development won't fix the risk in front of us. Frontier and open weight models are literally everywhere including in the hands of the adversary.

Kurtz frames the shift from security to AI safety as an accelerator for enterprises rather than a brake.

Security was keep everyone out. Safety is allow our organizations to go faster. And safety isn't the brakes. It's what allows you to run agents and go faster.

Kurtz believes token maxing is ending and open-weight models will drive cost reduction and data sovereignty.

I think all the token maxing might be over. And in fact, now we're hearing more and more on how do we reduce cost. And in my view, open weight models are going to be one of the key areas to help reduce cost and drive this data sovereignty.

Kurtz predicts a massive increase in open-weight models running locally on endpoints.

So here's my prediction. I think there's going to be a massive increase in openw weight models running directly on the endpoint. You can run a 28 billion parameter Quinn model on your desktop.

Kurtz argues general-purpose AI refuses defender tasks, so purpose-built defender AI is needed, motivating CrowdStrike's lab and models.

So the best AI on earth was built for everyone except the defenders for everyone in this audience. General purpose models are built for general purpose and we need everyone in this room needs AI built for the defender.

Kurtz warned that AI without security hands adversaries an opportunity and that security must be built into AI infrastructure from the start.

“Every enterprise is racing to build AI, but AI without security is an opportunity for the adversary,” Kurtz said. VIEW ALL “Security cannot be an afterthought in an AI-powered world.

Kurtz said Falcon's AI-native protection becomes foundational to the Dell and NVIDIA AI Factory, letting enterprises innovate safely at scale.

Dell and NVIDIA are powering the AI revolution, and CrowdStrike is securing it. Together, we’re making Falcon’s AI-native protection foundational to the AI Factory, giving enterprises the confidence to safely innovate at scale on validated, optimised architectures.

Kurtz said every enterprise will run on AI and that securing AI is CrowdStrike's largest market opportunity ever.

Kurtz said every enterprise will run on AI and that securing it is the largest market opportunity in the company’s history.

George Kurtz warns that AI without security gives adversaries an opening and that security must be built in from the start.

“Every enterprise is racing to build AI, but AI without security is an opportunity for the adversary,” said George Kurtz, CEO and founder of CrowdStrike. “Security cannot be an afterthought in an AI-powered world.

Prince says CISOs are panicking and the worst of the vulnerability wave is not yet over.

I think right now what CISOs are doing is panic. It feels like the sky is falling. Again, I think everyone is sort of holding their breath and saying, "Hopefully the worst of it's over." I know for a fact the worst of it is not over.

Prince says customers are buying the unified Cloudflare platform, turning point-solution Zero Trust vendors into mere features.

people aren't buying just Zero Trust. They're not just buying WAF, they're not just buying DDoS, they're not just buying the developer platform. They're buying the overall platform of Cloudflare, and they're seeing that if you incorporate and use all of what we do, that you get a much more holistic solution. As a result, the Zero Trust vendors effectively just become features within what is a much larger platform play.

Prince warns that a wave of severe vulnerabilities is coming and will persist for two more years.

Security: Six of the most severe vulnerabilities in history were disclosed within 14 days. The worst is yet to come, and this state will persist for another two years.

Prince argues first-generation Zero Trust vendors fail on agent identity and are being reduced to features as customers buy platforms.

First-generation Zero Trust vendors aren't addressing the issue of agency identity; clients are abandoning ongoing RFPs; customers are buying an "entire platform," and first-generation vendors are becoming mere features.

Matthew Prince says he has lived in abject terror over AI-found bugs leaking before patches, and predicts a critical internet-wide vulnerability roughly every week for the next two years.

“The last two months, I’ve been living in a state of abject terror that someone else would find the bug or the information would leak before everyone had patches and fixes in place,” said Matthew Prince, CEO of Cloudflare. “For the next two years, we’re going to be living in a state where, on average, every week there is a critical, internet-wide vulnerability.”

Prince argues that although the next two years will feel like the sky is falling, finding these flaws with AI will ultimately yield a much safer internet.

“You’re actually safer in a world where we are finding these things,” Prince said. “The next two years, it’s going to feel like the sky is falling, but I think we come out the other end of it with a much safer internet.”

Krebs argues that residential proxy services are a major source of malicious traffic, including account takeover, ad fraud and AI-driven scraping.

As detailed in several stories here over the past year, residential proxy services are a major source of unwanted and often malicious traffic, including account takeover activity, advertising fraud, and mass content scraping for AI projects.

Schneier warns that lawmakers will need to harden legislation, possibly using advanced AI themselves, as AI-driven legal hacking is definitely coming.

In such a world, lawmakers would need to significantly harden the process of legislation, for example by leveraging advanced AI themselves. Still in its early stages, but definitely something that’s coming.

Schneier highlights that AI-driven vulnerability discovery is finding critical software bugs at an astonishing pace, on track to add about 26% to the all-time total this year, and calls this staggering.

3% of ‘critical’ software security bugs, in all of recorded history, have been found in the last month. This is on pace to find another ~100,000 bugs this year alone, adding ~26% more vulnerabilities to the all time total Do you realize how fucking insane that is?

Zafrir argues Check Point is moving beyond firewalls toward preemptive security because the current security paradigm is showing holes.

we're no longer a firewall company, and we're no longer a network security company. We are looking at where the attackers are going, and I personally believe after 30 years in this industry that the current paradigm, we're seeing holes in the current paradigm, and I think it won't be long before many of the things that we were accustomed to will become less and less relevant. We have to move from looking at the rearview mirror, and we have to start anticipating attacks.

Zafrir expects AI data center buildouts to create incremental security demand for Check Point starting in 2027.

The investment in data centers, and specifically AI data centers, whether it's for training or for inference, is just mind-boggling. These data centers need security. In fact, that's some of the places where we believe this is going actually to be incremental starting 2027.

Zafrir says data sovereignty and local open-weight models are redefining network security and boosting on-prem demand.

I do think that network security, modern network security does not necessarily mean what it meant a couple of years ago. It also means a sovereign data center. It also means running open-weight models locally, sometimes because of sovereignty, sometimes because of the need to remain independent.

Zafrir warns of a 'vulnerability apocalypse' and says reactive security will lose, requiring proactive mitigation.

what we are seeing is a vulnerability apocalypse, if you like. The only way to stay ahead of that is to be able to thin the stacks and proactively, on the one hand, dramatically lower the probability that it will happen.

Zafrir calls security for AI almost existential for customers and Check Point, and says investment will be doubled next year.

Security for AI is one example of a capability or a product that we believe is almost existential for our customers, and hence for us.

Zafrir defends the 12-20% firewall price increase as reasonable and not harmful to demand.

We believe that it's a very reasonable price increase that won't hurt the demand.

Zafrir says demand is rising and the slower recovery stems from Check Point's own execution and sales hiring pace.

We actually see demand rising. Now, it is on us to be more competitive, to be out there, and to execute better.