Over 30 individual Minnesota water and wastewater treatment facilities were simultaneously hit with a cyber-attack from an unknown source on July 26 and 27, 2026. The coordinated attack targeted the utilities’ operational technology systems and caused some affected communities to request that residents minimize water use due to limited stored water. Other communities experienced equipment malfunctions requiring them to implement contingency plans and switch to manual operations.

The Minnesota IT Services agency activated its incident response plan statewide in response to the attack. Although the attacker is unknown, there is speculation that it may be attributable to Iran-backed hackers in response to attacks on a southern Iranian water treatment plant.

Australia and the U.S. Cybersecurity & Infrastructure Security Agency (CISA) issued guidance on July 28, 2026, for critical infrastructure operators to isolate vital systems during cyber attacks “or periods of increased cyber threat.” CISA has been warning critical infrastructure operators about increased threats from Iranian-affiliated cyber actors repeatedly since the war in Iran commenced.

The fact that the threat actors coordinated this attack to affect multiple utilities across an entire state is rather frightening. Critical infrastructure operators should stay informed of the CISA issued guidance and take the warnings seriously.

Sony smart TV owners have voluntarily dropped their proposed class action against Samba TV, an analytics company accused of collecting and selling television-viewing information to third-party advertisers in violation of state and federal privacy laws. DellaSalla v. Samba TV, Inc., No. 3:25-cv-03470 (N.D. Cal. 7/23/26).The dismissal came after the federal court had already allowed several claims to proceed, including intrusion upon seclusion, unjust enrichment, and claims under the Federal Wiretap Act and the California Invasion of Privacy Act (CIPA).

The plaintiffs alleged that Samba TV technology embedded in Sony televisions intercepted unique identifiers associated with their TVs and private video-viewing data without consent. Earlier in the case, District Court Judge Jacqueline Scott Corley found that allegations that Samba TV collected and sold detailed video-viewing information tied to political leanings and other private characteristics were enough to establish federal standing.

Even though this case has been dropped, the court’s earlier ruling remains important regarding the privacy risks connected device data, viewing data, device identifiers, ad-tech integrations, and inferred sensitive attributes can present when companies do not have clear consent flows, accurate disclosures, and tight controls over third-party data sharing. Companies using smart-device analytics, pixels, SDKs, automatic content recognition, or cross-device advertising tools should review what data is collected, whether it is linked to households or individuals, how consent is obtained, and whether vendor contracts and public disclosures match the technical reality.

On July 20, 2026, Pennsylvania Governor Josh Shapiro signed SB 992, updating and expanding Pennsylvania’s Telemarketer Registration Act of 1996 and strengthening restrictions on unwanted telemarketing communications. The law reflects that telemarketing is no longer limited to live calls and that texts, prerecorded messages, and other automated tools are increasingly reaching consumers and businesses .

The bill broadens “telephone solicitation” to cover traditional calls, voicemails, ringless voicemails, and text messages sent to residential, business, or wireless subscribers for sales-solicitation purposes or to obtain information for a future solicitation. It also expands who may qualify as a “telemarketer” to include persons or businesses that initiate or receive calls or messages involving Pennsylvania subscribers, and updates “robocall” to include certain automated solicitations using prerecorded or artificial voice calls or messages.

The law also creates a prior express written consent framework for robocalls and text messages. That consent must be documented in a written agreement that identifies the number that may be contacted, clearly discloses the consumer’s agreement to receive solicitations, states that consent is not required as a condition of purchase, and is signed electronically or otherwise. Subject to certain exceptions, SB 992 prohibits robocalls to residential, business, or wireless lines without that consent.

The bill also adds operational restrictions as well. Telemarketers may not place telephone solicitations on Sundays, legal holidays, before 9 a.m., or after 7 p.m. They also must promptly identify the call’s purpose, the telemarketer or telemarketing business, and, if applicable, the offer. For text solicitations, recipients may opt out by replying with terms such as “STOP” or “UNSUBSCRIBE.”

The law extends do-not-call protections to business and wireless subscribers and requires telemarketers to obtain applicable Pennsylvania do-not-call listings quarterly or use a service provider that does so. It also targets deceptive technology use, including improper consent practices and synthetic or computer-generated messaging that defrauds, deceives, or misleads recipients.

Violations may carry civil penalties of up to $1,000, or up to $3,000 if the person contacted is age 60 or older. Although the new law does not take effect until October, businesses using outbound calls, texts, prerecorded messages, ringless voicemail, or telemarketing vendors should review consent language, opt-out processes, suppression-list practices, and vendor oversight for Pennsylvania-directed telemarketing activity.

The White House is moving closer to a voluntary framework under which AI companies would submit their most advanced models to the federal government before public release. The White House’s Office of the National Cyber Director reportedly circulated the draft framework by to OpenAI, Anthropic, and Google, and those companies jointly submitted edits. Although the review process details are not yet public, the continued federal interest in pre-release review of frontier AI models follows earlier discussion of a possible FINRA-like watchdog for advanced AI systems. 

For businesses, the key takeaway is that voluntary AI governance is increasingly becoming a practical expectation, even where formal legal mandates remain unsettled. Companies developing, deploying, or procuring AI tools should be prepared to document model governance, risk assessment, testing, security controls, data provenance, privacy considerations, and human oversight in a way that can stand up to regulator, customer, investor, and board scrutiny. Even if the initial federal framework applies most directly to major AI model developers, downstream users should expect those norms to flow through vendor diligence, contract terms, audit rights, procurement questionnaires, and enterprise AI policies.

Business clients should use this moment to get their AI governance house in order: inventory AI systems and vendors, classify higher-risk use cases, update privacy and security reviews for AI-enabled tools, and build clear internal approval processes before rolling out new tools . For companies buying AI products, contracts should address testing, transparency, cybersecurity, data use restrictions, confidentiality, model training rights, regulatory cooperation, incident notice, and responsibility allocation if the tool produces harmful or noncompliant outputs. The practical step now is not to wait for a final federal rule, but to establish a defensible governance record that reflects a thoughtful assessment of risks, sound decision-making processes, and clear oversight and explanations of AI system management.

A recent Electronic Frontier Foundation (EFF) article reports that companies that make wearable trackers, whether for fitness or health, “are behind the times when it comes to basic privacy practices and transparency” and urges those companies to “improve those practices.”

The article, Most Smart Watches, Rings, and Bands Lack Basic Transparency Reports and Key Privacy Features, states that up to 40% of individuals in the U.S. own a wearable health device. These devices are not subject to health-related privacy protections, such as HIPAA. They can, and do, collect vast amounts of data while being worn, and the companies share that data with third parties or useit to train AI models. When users activate the device, its privacy policy outlines what information it collects, how it is used and disclosed, and who it is shared with. The user is asked to click “I agree” to the privacy policy, and unfortunately, most people don’t read it before they agree. This allows the device to collect, use, and share your data broadly, including with law enforcement.

EFF staff analyzed ten different companies that market consumer health products and found that most of them did not provide transparency reports, which provide information to consumers about how often they provide data to the government in an official or unofficial manner. Only Apple and Google publish transparency reports, and Apple, Google, and Whoop agree to notify users of law enforcement requests. According to EFF, Oura is reviewing its practices to “to provide greater visibility into how we handle these requests,” and Suunto is evaluating its transparency practices as well. EFF concludes that:

Any company that handles data that’s of interest to law enforcement and governments owes it to their users to publish transparency reports and, when legally possible, notify users when that data is requested. This is especially true of personal health data, which can reveal our movements, and be used to infer details about what we’re doing at any given moment.

The analysis also looked at whether the wearable device companies use end-to-end encryption for health data storage . It found that the Apple Watch “is the only popular fitness wearable that supports end-to-end encryption, and it’s enabled by default for all users (you are required to have two-factor authentication enabled as well, but that is also on by default for most accounts).” EFF warns users that this is only for the data stored in the Apple Health app, and not for other apps downloaded on the watch, or that are shared with other third-party apps or wearables. EFF found that Apple was the only one of all the companies reviewed, which “means those companies see and use your data,”…which “is a major privacy oversight, especially when you consider these devices collect heart rate, track sleep, and can log your location while also calculating a variety of health metrics supposedly intuiting everything from anxiety to your fitness ‘age.’”

EFF advises users to reach out to their fitness wearable to request that they offer transparency reports and provide end-to-end encryption for data storage. It also urges wearable companies to update their security practices and provide more transparency.

Cybersecurity firm Forrester recently issued its annual report, Top Cybersecurity Threats in 2026, which outlines “the most critical risks organizations need to plan for.”

The report predicts the top threats that organizations will face in 2026 based on recent trends and observations. The top five threats expected in 2026 include:

  1. Near autonomous attacks from a nation-state.
  2. Concerns over agent threats.
  3. Non-negotiable AI software supply chain.
  4. Provenance and IAM risks of AI agents.
  5. Digital sovereignty spans regions and tech stacks.

The threats listed above stem from the widespread accessibility of AI models and the ability for AI tools and AI agents to ingest vast amounts of data, learn from it, and operate at unprecedented speed and scale. These capabilities enable the automation and expansion of increasingly sophisticated cybersecurity attacks, the use of shadow AI beyond an organization’s governance and visibility, and heightened supply chain risks. Together, these challenges create significant compliance concerns, reduce organizational control, and increase overall cybersecurity exposure.

The report outlines the most significant risks organizations face in 2026,  including the growing challenges posed by AI, and underscores the importance of implementing a robust AI governance program as a critical strategic priority

On September 10, 2025, the U.S. Department of Defense (DoD) issued the CMMC Procurement Rule, which made cybersecurity compliance a condition of doing business with that agency by requiring contractors and subcontractors to meet specified security standards before accessing Federal Contract Information (FCI) or Controlled Unclassified Information (CUI). We previously covered the CMMC Procurement Rule and its requirements in detail here.

Less than a year later, on July 13, 2026, the DoD issued guidance pausing the next scheduled step in the CMMC rollout, which had been expected in November 2026 and would have expanded the use of more formal third-party and government-led assessments. The DoD’s memorandum ties the pause to DoD’s broader effort to reduce acquisition process, speed delivery of new capabilities, and avoid placing unnecessary burdens on small and non-traditional businesses in the Defense Industrial Base.

Contractors should understand that the CMMC Phase 2 suspension changes the assessment process, not the underlying cybersecurity obligations. In addition, the “Phase 2” pause should not be confused with CMMC Levels 1 and 2. The phases refer to the DoD’s rollout schedule, while the levels refer to the type and sensitivity of information a contractor handles and the corresponding cybersecurity requirements. During the suspension, program managers and requiring activities may include only CMMC Level 1 self-assessments or CMMC Level 2 self-assessments in procurement documents. They may not require Level 2 third-party assessments by a C3PAO or Level 3 assessments by the Defense Industrial Base Cybersecurity Assessment Center (DIBCAC) during this period.

Even so, the DoD stated that DFARS 252.204-7012 remains in effect, and that baseline compliance with NIST SP 800-171 Rev. 2 will continue to be enforced through self-assessments and select government-led assessments. That means federal contractors still need to be able to show that they are protecting federal information properly. Level 1 applies to FCI, such as non-public information provided by or generated for the government under a contract, and is tied to FAR 52.204-21, which sets basic safeguarding requirements for contractor information systems, such as limiting system access to authorized users, controlling physical access to systems, and using basic protections against malicious code. Level 2 applies where CUI is involved, such as technical drawings, specifications, or other sensitive government information that requires safeguarding, and remains aligned with NIST SP 800-171 Rev. 2, a more detailed set of security controls for protecting that information, including requirements for access control, incident response, system monitoring, and security assessment.

The guidance also affects live procurements. If a solicitation or requirements package included a Level 2 C3PAO or Level 3 DIBCAC requirement, DoD personnel must initiate amendments, and contracting officers or agreements officers must issue corresponding solicitation amendments “as soon as practicable.” Existing contracts or agreements containing those requirements are also to be modified.

Federal contractors should monitor solicitations, amendments, and contract modifications closely, but they should not pause cybersecurity work. The DoD’s Chief Information Officer is conducting a 60-day review of CMMC to ensure the Defense Industrial Base remains secure without imposing significant burdens on small and non-traditional businesses. Until the DoD issues further guidance after that review, contractors should keep policies, system security plans, plans of action, and self-assessment records current and supportable.  

California’s privacy regulator has launched its first-ever audit, signaling a new phase of active oversight under the California Consumer Privacy Act (CCPA) and its amendments. The California Privacy Protection Agency (CPPA) is focusing on delivery and transportation apps in the gig economy, examining how platforms collect and use personal information from both consumers and workers, and how individuals can exercise rights to know what data is collected, how it is used, and with whom it is shared.

The agency’s choice of sector is notable. Gig economy platforms (i.e., food delivery and transportation apps) often rely on highly sensitive and operationally important data, including geolocation data, behavioral and performance metrics, biometric data, communications records, and other information that may be used to make decisions about assignments, ratings, compensation, and account status. According to the CPPA, hundreds of consumer complaints and public comments during rulemaking in part prompted the audit by with officials noting particular concern about the rapid evolution of employee monitoring technologies and AI-enabled data practices.

As part of the audit, the CPPA plans to review platform policies and practices, request documentation and data, interview company personnel, and directly test app processes. The agency‘s inquiries will be narrowly focused and it plans to publish an industry compliance report similar to the Federal Trade Commission’s Rule 6(b) market studies. For companies operating in or adjacent to the gig economy, the audit is a reminder to pressure-test privacy notices, rights-request workflows, data sharing disclosures, employee and contractor privacy practices, retention schedules, and governance around sensitive data and automated decision-making before regulators come knocking.

California’s SB 361 expands California’s Delete Act and will soon require registered data brokers to regularly check California’s data deletion database, known as DROP, to determine whether a California consumer has requested deletion of their personal information. Beginning August 1, 2026, data brokers must access DROP at least every 45 days and, when a request appears, delete the consumer’s personal information within 45 days and direct applicable service providers and contractors to do the same. Once the information is deleted, the data broker generally may not sell or share new personal information about that consumer unless the consumer indicates otherwise.

Companies that collect, buy, sell, or share Californians’ personal information should carefully evaluate whether they fall under California’s broad definition of a “data broker.” In general, a data broker is a business that knowingly collects and sells personal information to third parties about consumers with whom tit does not have a direct relationship. This can include businesses outside California, and the analysis may be more complicated for companies using third-party tracking technologies, purchasing personal information from others, or selling data collected indirectly. Data brokers also must register annually with the California Privacy Protection Agency between January 1 and January 31 and provide detailed disclosures, including categories of personal information collected and whether personal information is sold or shared with certain recipients, such as generative AI developers.

The penalties are significant: failures to register can trigger $200-per-day penalties, and failures to honor deletion requests can result in $200 per deletion request for each day the information is not deleted. Companies that may fall under California’s definition of a “data broker” should prepare now by confirming whether registration is required, mapping California personal information flows, identifying vendors and contractors that may need deletion instructions, and building a documented process to check DROP at least every 45 days. Even companies that do not intend to collect information from California consumers should revisit their data sources and screening practices, as the CPPA has already shown interest in out-of-state businesses that handle Californians’ personal information.

A new study by the AI Security Institute (AISI), Cheating Behaviour in Frontier Model Evaluation, found “cheating behaviour in all of our capability evaluations,” and outlines “the implications as models grow more capable.”

AISI defined “cheating” as “taking an action that is out of scope for the task or explicitly disallowed by the rules, in order to achieve a goal through a shortcut, workaround, or unintended solution that the task was not meant to, or should not, permit.” Sounds like cheating to me.

AISI found that “every model” it tested “attempted to cheat.” Even when called out on it, the models “did not reliably report this behaviour when asked, and often did not reason about it in their chain-of-thoughts, suggesting that detecting cheating will likely require robust monitoring methods.” That said, AISI notes that based on its review, to the best of its knowledge, “no model has successfully cheated (i.e. not been caught) in the results we report.”

To test the models, AISI tasked the models with evaluating cybersecurity capabilities, with specific, controlled and defined scope and tasks. It found that every model tested “cheated,” going “outside the scope or takes an action that the rules explicitly prohibit.” Importantly, the models were not asked to cheat or go outside the tasks and scope—they did so on their own.

When AISI asked the models if they went outside the scope or tasks, they only acknowledged the behavior and “described it as wrong” less than 50% of the time. AISI concludes that asking a model if it is cheating or going beyond the tasks assigned is not an effective way to monitor output. It also found that cheating is not in a model’s chain-of-thought reasoning, and therefore, this too, is an “insufficient method” for detecting cheating.

Why do we care if a model is cheating? The most obvious one is that if models go beyond specified tasks and scope, cheating can be “especially dangerous in domains where verifying success is hard, such as AI safety and security research, or where the cost of unintended actions may be very high, such as cyber operations or military decision-making.” It can also affect the efficiency of the use of the AI models if additional verification is needed. Finally, as models become more capable, their cheating behavior may become harder to detect and “more damaging when successful.”

AISI posits that models should not be trusted when they self-declare that they are not cheating. They suggest that users monitor and detect cheating by “combining manual review with additional tools such as the LLM monitor.” As models become more capable, “a more fundamental fix would be to train the models not to cheat in the first place.” Those developing AI models should take this recommendation to heart now, before AI models train themselves to evade cheating behavior detection. Train AI models not to cheat now, during development, so the problem does not become worse.