CrowdStrike announced Falcon Guardian, a product in the AI detection and response (AIDR) category. It extends the company’s endpoint-detection approach for humans and computers to AI agents, which can access data, computing and networking resources.

George Kurtz said CrowdStrike developed the technology with large customers, including Amazon. He characterized security as a potential accelerator rather than a brake for AI adoption, and said the company is applying roughly 15 years of accumulated data to its work on generative AI and agents.

CrowdStrike and NVIDIA announced SafeMind, a cybersecurity-focused AI initiative built on NVIDIA’s Nimotron. The system combines specialized models with multiple harnesses designed to reason, call tools, and work through attack-and-defense tasks such as finding vulnerabilities and writing detections.

CrowdStrike said it is deploying SafeMind internally and building a digital twin of its environment for attack-defense simulations. The companies said their goal is to make the system better, faster, and more cost-effective than other open-source and frontier models for cybersecurity use cases; the claimed Arc AGI 3 benchmark improvement from 30% to 100% was presented by the companies and was not independently verified in the discussion.

The SafeMind system includes the Red Tempest offensive model and Blue Solano defensive model. CrowdStrike said both the models and harnesses will be available through its platform, with direct customer access offered through a trusted-access program in Quiltworks. The companies also described a red-blue training loop in which defensive models continually learn from offensive models.

Kurtz said CrowdStrike’s security system links indicators of attack across known attack chains and uses AI to connect signals and remove noise. He said models perform well at separating relevant signals from large volumes of data.

He distinguished this analysis from immediate breach prevention. The Hugging Face incident was described as a forensic question of what happened, while deterministic security must operate inline and make the right decision the first time. Kurtz said he had not seen models stop a breach as it was occurring, adding that they are useful for sorting through data, finding vulnerabilities and determining what happened.

Kurtz also said models such as Mythos had not invented a new hacking technique. In his view, they make existing techniques more scalable and faster by finding and linking more vulnerabilities, rather than creating a wholly new form of attack.

Nation-states, e-crime groups and hacktivists are leveraging open-weight models, CrowdStrike says. Models with their guardrails removed can be downloaded from Hugging Face, run on a powerful system and, when prompted, generate a malware or ransomware kit.

CrowdStrike characterizes the capability as an immediate security issue rather than a theoretical one. Customers want visibility into agents and tools to stop these threats, while determining whether an incident is a fully AI-driven attack or an adversary using AI assistance.

The company says that distinction informs both defensive measures and how an incident is communicated to executives and boards. CrowdStrike says its existing agent and platform can instrument agent actions, identities, tool calls, agent spawning, network connections and prompts.

CrowdStrike says Falcon Flex is a token-based licensing system for AI-related security use: customers can draw down credits and pay more as their AI consumption increases. The company says it uses multiple models to reduce costs and aims to preserve customers’ data and data provenance.

CrowdStrike also says organizations are entering 2026 with large security requirements but constrained budgets as they seek tokens to fund AI activity. One company representative described “token shock” as unexpectedly high spending, citing an example of $20,000 more than expected per employee per month; the transcript does not establish whether that example is typical or independently verified.

The company presents Flex as a more efficient procurement vehicle, emphasizing not only purchase cost but also deployment and day-to-day operationalization of cybersecurity technology.

CrowdStrike says board and CEO attention to AI-related threats has increased. Fortune 500 companies generally have, or can find, funding for cybersecurity, although spending varies by industry and regulation.

Local utilities, hospital systems and NGOs are identified as organizations that may have fewer resources. Some operate older software and lack enough security personnel, advanced software and other technologies.

CrowdStrike says helping these organizations requires a collective community effort, including work with OpenAI and other partners. No specific assistance program, funding amount or delivery mechanism was identified.

CrowdStrike says its Superintelligence Lab has 270 PhDs and that AI use has expanded across the company, with coding expected to be a major area. The company is also deploying AI builders across different organizational functions.

CrowdStrike says generated code must be checked for security. It warns that AI can reproduce vulnerabilities in code and that a model may not reliably assess the security of its own output.

The company says employees who reject AI because they fear for their jobs will not be successful at CrowdStrike. Those who use AI in the right places, at the right time and at the right cost have “all the room in the world” to succeed, it says.

CrowdStrike president Michael Centonas said go-to-market, product and engineering, research, and threat-hunting teams report to him. He said he previously served as the company’s CTO and works closely with its engineering and research leadership.

Centonas said strong technology is insufficient if it cannot be sold, its value communicated, or deployed by customers to help them stay safe. He described his role as bringing those functions together to keep customers secure.

Cybersecurity defenders face a structural disadvantage: they must prevent attacks consistently while keeping normal business operations running. Unlike attackers, they are constrained by change-control procedures, regulatory requirements, privacy rules and limited IT and token budgets.

The analysis argues that adoption speed and access to open-weight models may matter as much as raw model intelligence. Adversaries can download, customize, train and privately run models, making their use difficult to detect until an attack occurs. The long-predicted point at which attackers weaponize vulnerabilities faster than defenders can respond is described as having arrived, although the level of compute available varies from constrained individuals to well-funded nation-state actors.

Governments are discussing how to use AI while addressing risks from malicious actors, including the potential use of open-weight models. The cybersecurity focus is expanding beyond Fortune 500 companies to small businesses, which often lack dedicated security personnel, resources, budget and clear access to expert help.

The discussion emphasizes protecting critical infrastructure and essential services—including power, water, waste services, banks, hospitals and utilities—so economies can keep operating and banks are not compromised by ransomware. Community-wide collaboration and government partnership are described as essential; without them, governments and organizations may struggle to keep up with AI-enabled threats.

Attackers may exfiltrate encrypted data they cannot use today, potentially hoping that future quantum computers will enable its decryption. One scenario raised was that data described as encrypted could become readable in five years, but the precise purpose of large-scale theft may be unclear when it is discovered.

When attackers take terabytes of data without immediately selling or using it, the analysis suggests the activity still reflects an underlying motive. Possible explanations include delayed decryption or training models, making quantum computing a security issue that warrants some attention alongside AI.

Attackers are using ChatGPT and Gemini to write polished phishing emails, weakening poor grammar and unnatural language as traditional warning signs. A cited account said phishing click-through rates rose from about 11–12% roughly a year ago to more than 60%, attributing the increase to messages that appear highly realistic and grammatically polished.

The account also said open-weight models can provide a playbook for carrying out an attack, lowering the expertise barrier for sophisticated cyberattacks. It described AI-generated phishing as increasingly difficult to distinguish from legitimate communications, while presenting the lower barrier as a risk that more people will be able to conduct such attacks.

CrowdStrike’s leadership says the company’s response to market volatility is to keep putting customers first, continue innovating, and focus on keeping them safe and secure. It describes that principle as a company-wide mantra.

The company rejects the idea that AI systems finding vulnerabilities will create a stable period with fewer security problems, calling the “SaaSpocalypse” thesis meritless. Instead, it expects more vulnerabilities and attacks, making cybersecurity increasingly difficult.

CrowdStrike attributes this view to its experience and to an engineering team at the forefront of the field, alongside its continued customer focus.

Nick Carter posted that many people were receiving unsolicited X password-reset attempts in their email inboxes. The reports do not establish who initiated the attempts or whether they were part of a coordinated attack.

X users were advised to enable password-reset protection under Settings → Security and account access → Security check → Password reset protection. The activity was associated with a possible financial incentive as X Money rolled out to more people, but that connection was not established as fact.

YouTube creator and director Markiplier has acquired an 8.5% stake in GoPro. The action-camera maker has entered a definitive agreement to merge with privately held Starman Optical in a transaction valued at $285 million.

Starman Optical describes itself as a U.S. optical-photonics company focused on developing and domestically manufacturing optical transceivers. Markiplier’s specific strategic role remains unclear, though possible directions include promoting new GoPro products or repositioning the brand toward filmmakers and cinematic production. The deal comes as GoPro faces competition from smartphones, drones, Insta360 and DJI.

Meta’s Oakley Meta smart glasses are emerging as a consumer capture device competing with action cameras such as GoPro, while products from Insta360 and DJI remain part of the comparison.

Best Buy reportedly said smart glasses were driving significant in-store sales and selling well. The reported retail momentum has not been independently observed by the sources, who noted they had not personally seen the trend.

MrBeast launched The Most Dangerous Games, a mystery novel written in collaboration with James Patterson. The book was promoted across MrBeast’s channels with a chance for one purchaser to win $1 million, effectively making the launch a lottery-style promotion.

Two people described as close to the project reportedly said preorders were in the four-digit range as of the prior week. That figure was not independently confirmed, and the promotion’s timing and details were described as uncertain.

The excerpts also note that raffle-style promotions may allow entry without buying the book, but they do not establish whether that option applies to this specific campaign.

An accelerated-computing company says different AI models and stages—including prefill, inference and decode—require different capabilities. Its strategy is to offer high-performance architectures across those workloads, alongside multiple rack-scale server configurations; a Groq deal is also coming online, although its precise future role was not specified.

The company describes rack-scale infrastructure built around seven processors, aiming to achieve best-in-class performance across them and maximize token efficiency per watt for large AI factories. It also sees a partner ecosystem as a way to extend that efficiency into new domains and use cases.

The future mix of specialized accelerators and configurations remains unsettled, but the described approach suggests that the fleet may expand as new models and use cases emerge.

John Ternus has appeared on X, although he has not yet used the platform to publish substantial news. His post was described as a lowercase “hello.”

The commentary characterized X as a place where the AI conversation is actively taking place, noting that Jensen Huang and Mark Zuckerberg are also on the platform. It expressed the view that people in important business positions can no longer afford to be absent from X, while leaving uncertain whether Ternus’s presence will lead to future news or more significant communication.

A Wall Street Journal analysis by Ralph Winkler says Apple investors expect incoming CEO John Ternus to be a stronger innovator, particularly in artificial intelligence. Tim Cook is credited with scaling Apple: annual iPhone sales rose from 72 million when he took over to a cited 255 million, while the company returned more than $1 trillion to shareholders.

The analysis says operational efficiency is no longer enough and that Ternus should launch new products and take greater risks. Apple trades at about 33 times next year’s earnings versus 20 times for the S&P 500, despite earnings growing half as fast as the market; investors are paying a premium for perceived safety amid uncertainty over returns on AI spending. The analysis argues that this valuation could become vulnerable if Apple fails to deliver innovation, while it may already have lost its former role in defining how consumers interact with devices. What Ternus will do remains uncertain; the analysis expresses concern about an apparent retreat from Vision Pro.

Dyson has released an AI-powered toothbrush with an integrated 100,000-pixel macro-lens camera, priced at $499.

The camera’s resolution is equivalent to about 0.1 megapixels. The practical value of adding a camera and AI functions to a toothbrush remains uncertain, with doubts raised about what the 100,000 pixels would be used for.

A final assessment was deferred pending testing of the device.