My buddies at CrowdStrike have uncovered a campaign launched by threat actor VAPOR PANDA starting on October 17, 2026, where ARTEX (an AI-powered penetration testing tool developed in China) is used with the Chinese-developed DeepSeek LLM to attack “more than 200 distinct organizations worldwide, including foreign ministries in Latin America, Europe, Asia, and the Middle East; semiconductor, defense, and think tank organizations in the U.S.; and telecommunications providers in multiple regions.”

According to the CrowdStrike researchers, they observed an ARTEX controller linked to VENDOR PANDA infrastructure direct exploitation activity “against entities in at least 14 countries over an 11-day window.” This included “19 U.S.-based organizations in the pharmaceutical, semiconductor, defense, nuclear, and financial sectors.” CrowdStrike confirmed that VENDOR PANDA was able to gain access, deploy malware, conduct credential spraying, network scanning, and reconnaissance.

What was different about this campaign was the speed and scope of the operation. In this campaign, the AI agents used the pen testing tools to run multiple engagements that allowed the agents to pursue “multiple attack paths in parallel, spanning web applications, cloud services, container environments, Active Directory, and mail and VPN infrastructure. They shifted to alternative approaches when a path failed. This breadth likely enables a single operator to apply a wider range of techniques against more targets at once than would be feasible through manual operations.”

The use of ARTEX allowed the threat actors to attack more targets at a faster speed than seen before. CrowdStrike posits that the use of ARTEX, paired with commercially available LLMs will allow “small teams to quickly scale intrusions across many targets simultaneously.”

CrowdStrike recommends:

  • Rotate signing keys, preauthentication keys, and service credentials after any suspected data exposure, including keys contained in historical backups and directory exports.
  • Enforce SAML signature validation, require IMDSv2, and restrict cloud roles’ command-execution permissions to the minimum required instances.
  • Hunt for ARTEX markers, including command-line parameters containing artx or artex strings, artx-prefixed cloud roles, art18- administration-command comments, and the C:\ProgramData\artx5v\ directory.
  • Patch internet-facing appliances, particularly long-lived Linux-based servers, and ensure they run the latest available software and firmware.
  • Deploy an advanced, real-time endpoint detection and response (EDR) agent, such as the CrowdStrike Falcon sensor, across the environment, including on servers considered inaccessible from the public internet.

CrowdStrike published the indicators of compromise, which should be reviewed and blocked by IT professionals.

On October 29, 2026, General Motors (GM) agreed to pay $220 million to 35 state Attorneys General over allegations that its Smart Driver program associated with its OnStar service violated the privacy of users. The AGs alleged that GM illegally sold drivers’ private location and driving behavior data to brokers without users’ consent, who then sold it to insurers that allegedly used it to set and increase auto insurance rates.

The complaint against GM included allegations that the data collected and sold included: “huge volumes of consumer driver behavior information, such as hard braking, hard acceleration, hard cornering, crossing of a designated high-speed threshold, seat belt usage, late night driving, and trip time and duration for such events.” Through the OnStar program, GM also collected “precise geolocation data linked to identifiable personal information, including name, address and Vehicle Identification Number.” The geolocation data included latitude and longitude, elevation, speed and a trip identifier that could tie together to “identify the entire route of a specific trip taken by a single vehicle.” Precise geolocation data was collected every three seconds from the time the user turned the vehicle on until the ignition was turned off. The complaint alleged that GM’s enrollment process failed to disclose the sale of their data and “contradicted its representations … that it would only share anonymous driving data with insurance companies. In reality, GM sold the identifiable driving data of all Smart Driver participants” to third parties for its repackaging and sale to insurance companies. The complaint alleges that GM made over $20 million from the sale of the data.

In addition to the settlement amount, the Consent Decree provides that GM agrees to limit the driving data that it collects, obtain express consent from drivers before collecting or sharing the data, develop, implement and maintain a comprehensive information privacy program for the collected driver data, including being transparent with drivers about the collection, use and sale of driver data, employ a Chief Privacy Officer, perform data protection assessments when engaging in a new high risk processing activity, implement data minimization efforts, provide consumers with a transparent privacy notice, and obtain express consent prior to collecting or using driver data. All of these items are solid measures to employ when collecting the type of data that is the subject matter of the settlement.

This settlement follows GM’s prior agreement to pay California $12.7 million to settle an investigation into its alleged sale of driver data to data brokers, and a January 2025 settlement with the Federal Trade Commission to refrain from disclosing vehicle geolocation and driver behavior data to consumer reporting agencies.

Drivers should be aware that their cars are collecting large amounts of very specific data about the functionality of the car, and the behavior of the driver. Be sure to read the privacy policies of the manufacturer so you are educated about the data collected, and used on your vehicle and your driving behavior, and make informed consent prior to the collection and use of that data.

On October 22, 2026, Pennsylvania Governor Shapiro signed the Artificial Intelligence Deceptive Advertising Act, setting disclosure rules for AI ads in the state.

The Act defines “synthetic content” broadly to include images, video, audio, or text significantly modified or generated by AI. It prohibits advertisers from knowingly, recklessly, or willfully creating or disseminating ads that use such content to create a false impression about a product’s or service’s nature, origin, or authenticity, unless the ad includes a clear and conspicuous disclosure.

The disclosure must identify the content as AI-modified or AI-generated, appear the first time consumers encounter it, and be provided in the same format as the synthetic content. Where technically feasible, it must also be “permanent or extraordinarily difficult to remove.” The disclosure must be understandable in the language or other interaction the consumer uses, consistent with the ad, and, depending on its format, audible or readable. There are several exemptions, including for fictional depictions, parody, satire, commentary, and criticism. Ads limited to text are also excluded from the disclosure requirement, even if the text was generated or significantly modified by AI.

The Act takes effect December 21, 2026. The Attorney General or a district attorney may seek an injunction or bring a civil action. A court may impose a civil fine of up to $3,000 for a violation.

For organizations using AI for advertising, the risk arises when an ad could mislead consumers. A synthetic product image could raise concerns if it misrepresents the offering, and a disclosure may still fall short if consumers are unlikely to notice it or if it conflicts with the ad’s message. Review will be especially important for realistic synthetic media that could be mistaken for a genuine product, person, or endorsement.

AI chatbots can feel like patient, empathetic companions, ready to explain a difficult subject or offer reassurance at any hour. However, a confident, supportive tone can make unreliable advice seem trustworthy, especially when a person is vulnerable or facing a serious health decision. That concern is at the center of a lawsuit filed by the parents of Bowen Wan, who allege that ChatGPT encouraged their daughter to reduce medication for severe epilepsy, conceal the change from her doctors, and dismiss potentially serious symptoms. Wan died in February 2026. OpenAI called the case heartbreaking and said it involved an older version of ChatGPT. The allegations have not been proven in court.

Research cited in that matter points to a broader design challenge: some AI systems may agree with users too readily, reinforcing mistaken beliefs rather than questioning them. Studies do not establish what caused Wan’s death, but they raise concerns about how a chatbot may behave over a long conversation, particularly when a user is seeking guidance about health or other high-stakes decisions. Other lawsuits involving ChatGPT and other AI platforms raise related questions about alleged harmful advice, self-harm, and emotional dependence. Each case turns on its own facts; a harmful outcome alone does not establish legal responsibility.

For AI companies, the challenge is to build systems that do more than sound helpful. That means testing how they respond to vulnerable users and extended conversations, recognizing when a situation calls for urgent human help, and pushing back on unsafe assumptions instead of simply affirming them. A disclaimer that a chatbot is not a doctor may matter, but it does not resolve every question about product design, warnings, or foreseeable risks. Chatbots can be useful tools, but users should not mistake warmth or confidence for professional judgment, and companies should be prepared to answer for the safeguards behind that trust.

This post was authored by Nathaniel Arden, partner in Robinson+Cole’s Health Law group + Aarathi Kallur, Legal Intern in the firm’s Health Law group. Aarathi is not permitted to practice law.

On October 13, the American Medical Association (AMA) released a governance framework for the use of agentic artificial intelligence (AI) that performs clinical functions in health care. The framework rests on a central principle: “[C]linical authority should be earned through evidence and maintained through demonstrated performance. Authority and oversight can evolve as evidence, performance, risk or the technology itself changes.” The AMA guidance is not binding on any health care providers, though it may provide a useful template as health care providers increasingly seek to implement agentic AI in their clinical practices.

The AMA’s framework outlines four topics and related questions for health care providers to ask while implementing agentic AI in clinical practice. The AMA’s framework and recommendations are summarized as follows:

  1. Function: What Clinical Function Does the AI Perform?
    The AMA suggests that clinicians should identify the AI’s task, intended use, patient population, and clinical setting. Functions may include documenting, monitoring, triaging, providing care recommendations, and diagnosing.
  2. Evidence and Risk: What Shows that the AI Can Perform the Function, and What Happens if it Fails?
    The AMA recommends that clinicians should evaluate the AI’s performance, safety, clinical benefit, population fit, and uncertainty. The rigor of that review should reflect the potential consequences of the AI’s failure.
  3. Authority and Oversight: What May the AI Do, and Under What Conditions?
    Under the AMA framework, providers should define the AI’s permitted scope of action and the human or institutional oversight required for its safe use.
  4. Accountability: Who Is Responsible for the AI’s Risks and Outcomes?
    Importantly, the AMA framework directs providers to assign accountability based on each participant’s authority, control, knowledge of the risks, and ability to prevent or mitigate harm.

The AMA recommends that steps 2 and 3 be frequently reviewed and refined through monitoring and reassessment.

The framework follows the AMA’s creation of the Center for Digital Health and AI, which seeks to help physicians shape how AI is developed, implemented, and regulated. That work has become more urgent as claims about AI’s clinical capabilities grow. The AMA’s agentic AI framework offers a practical standard to assist clinicians in adopting agentic AI together with appropriate planning, oversight, and clear accountability.

This post is also being shared on our Health Law Diagnosis blog. If you’re interested in getting updates on developments affecting health law organizations and policy, we invite you to subscribe to the blog.

Last week was Digital Parenting Week. I must admit that I was not aware it existed, but certainly believe it’s a good idea. Parents must talk to their kids about the risks of the digital world and how to protect themselves. I previously commented on tips to strengthen children’s privacy in a digital world in a previous Privacy Tip The post still resonates and is applicable today.

In honor of Digital Parenting Week, the Electronic Frontier Foundation (EFF) has published some tech tips to help parents with navigating discussions with their kids on how they “can work together to co-curate a digital safety plan that helps all members of the family enjoy the benefits of technology while protecting their online privacy and safety.” I am a big fan of EFF, and their suggestion that parents work together with their kids to talk about what they want to protect, who they want to protect it from, how bad are the consequences if we fail, how likely is it that we will need to protect it, how much trouble are we willing to go through to try to prevent potential consequences, and who are our allies, give parents a clear strategy on how to approach the issue in a non-threatening way. I highly recommend reading the piece to give ideas on how to start a conversation with kids about a tough subject-matter.

Security Company Arctic Wolf has reported that it has documented more than 400 incidents where threat actors flooded employees’ inboxes with spam, to the point of extreme frustration, then used Teams to contact the frustrated employee posting as IT support to assist in fixing the problem. The Teams contact comes within minutes of the flooding of the inbox. Because the Teams contact is from a legitimate platform and is not a vulnerability with Teams, the victim believes it is legitimate and accepts the help. This is another example of how threat actors use legitimate tools in a malicious way.

In using the legitimate Teams platform, the threat actor can bypass internal security measures to detect unauthorized activity, and legitimate IT professionals are unable to detect or defend users against the activity.

When the victim responds, Arctic Wolf found that it took the threat actor less than 15 minutes to obtain remote access. First, the threat actor asks the victim to switch to a Teams voice call and the threat actor then walks the victim through the installation of remote access software. The threat actor deploys the remote access software and starts running malicious code.

Be wary of any call or message from an outside organization, and only accept external chats with “specific partner domains.” IT professionals may wish to block “unmanaged Teams accounts and trial tenants and limiting Quick Assist and other remote access software to authorized IT staff.” Users should be aware of this threat and the unauthorized use of Teams, and to avoid any Teams chats or messages initiated from outside their organization. If your inbox gets flooded with spam, or you receive an unknown Teams call or chat, don’t answer it or respond–call your internal IT professionals on a reliable internal number.

On October 26, 2026, a federal judge denied ed-tech provider PowerSchool’s motion for partial judgment, keeping a California medical-privacy claim alive in In re PowerSchool Holdings, Inc. & PowerSchool Group, LLC Customer Security Breach Litig., No. 25-md-03149-AJB-MSB, slip op. at 12 (S.D. Cal. Oct. 26, 2026).

The case stems from a 2024 data breach that allegedly affected nearly 50 million K–12 students and teachers. PowerSchool provides software and data-management services to school districts. The plaintiffs allege that the company-maintained student medical information, including diagnoses, medications, and allergies, and made it available to school personnel and health-care providers. The court did not decide whether that handling violated the law. Here, it only decided whether the plaintiffs’ allegations were enough for the claim to continue.

You may wonder why an ed-tech company is facing a health-care privacy claim. Under California’s Confidentiality of Medical Information Act (CMIA), whether the law applies can depend on what a company does with medical information, not just the industry it serves. The CMIA generally bars covered entities from disclosing a patient’s medical information without authorization. To proceed, plaintiffs must plausibly allege that PowerSchool falls within a covered category, such as a health-care provider or contractor. The CMIA’s definition of a health-care provider can include a business that maintains medical information to make it available to individuals or providers for managing health information or for diagnosis and treatment.

PowerSchool argued that a recent California Supreme Court decision involving another education-technology company foreclosed the claim. The judge disagreed because that case involved allegations that student data was used for educational screening, while these plaintiffs allege that PowerSchool maintained and shared medical information for health-care purposes. According to the judge, that was enough for the CMIA-provider theory to proceed. A school-software company’s role can extend beyond education. When a platform handles student medical records to support care, it might find itself in the CMIA’s provider lane, too

AI notetakers are widely used by businesses in all industries. Although efficient, they pose privacy and cybersecurity risks, including lax compliance with established privacy and security frameworks such as GDPR or SOC 2, weak encryption, and recording conversations without consent, which may be in violation of one-party recording laws. For lawyers, it could mean an unauthorized waiver of privilege.

Before deploying an AI notetaker in your environment, it is important to assess the risk it poses to your organization, like any newly deployed technology. Unfortunately, many AI notetakers are embedded in conference platforms, and it may be difficult for users to turn them off. Consider establishing a process to approve AI notetaking for users, including training employees on which platforms have been sanctioned, obtaining consent from all participants during the use of the AI notetaker, establishing who has access to transcripts and how they are maintained, vetting vendors on training from your data, and reviewing vendors vendor contracts for privacy and security protections. Obvious places where AI notetaking may be prohibited include meetings with legal counsel, human resources meetings, and meetings discussing highly confidential matters.

Another consideration is how employees are using private AI notetakers in the workplace. Shadow AI notetakers are being used by employees in their personal lives, and that use spreads into the workplace. They can present themselves as apps on mobile phones or watches, or could be recording through sunglasses or wearable devices. The use of shadow AI notetakers presents a risk to the organization as the use is not sanctioned, governed or subject to the organization’s established policies and procedures.

Shadow AI notetakers present real risks to an organization as users may not be disclosing the use of the AI notetakers to others they are recording, are not obtaining consent, and are secretly recording conversations. In one example, an executive from an organization was having a private conversation with another employee, and unbeknownst to him, the other employee was secretly recording the conversation and the executive was fired.  This shows how ubiquitous AI notetakers have become and that there is no such thing as a private conversation anymore. Executives need to be aware of the risks associated with AI notetakers, assess them, and put a governance program in place to address their use in the organization like any other newly deployed technology. Employees need to be aware of the risks, follow the organization’s guidelines and ask that AI notetakers be dismissed from meetings discussing legal, human resources, or highly confidential matters.

Corporate legal departments are increasingly looking to see how outside counsel use artificial intelligence (AI). At a recent DRI, Association of Lawyers Defending Business conference, legal leaders from Bayer, Fifth Third Bank, and Progressive discussed how AI is entering outside-counsel guidelines and requests for proposals. One panelist captured the competitive shift: outside counsel who learn to use AI may replace those who do not.

This shift is part of a broader change in the work of in-house legal teams. The panelists described spending more time on emerging technology, including how to advise businesses on AI and how to incorporate it into internal operations. They also emphasized the value of adaptability as organizations respond to new demands. For clients, the question is not simply whether their lawyers use AI, but whether they can help make its use practical, thoughtful, and aligned with the business. Working with clients on the legal and governance considerations that arise as they adopt and use AI helps connect the technology’s potential with the responsibilities that come with it. As AI becomes part of the client-counsel conversation, that combination of practical experience and sound judgment will matter more and more.