X-Risk Daily

CISA, the NSA, the FBI, the Department of Energy and the Environmental Protection Agency co-authored the advisory, designated AA26-231A, to warn owners and operators of industrial control systems of an active cyber threat to Siemens S7 Series PLCs. The alert covers every generation of the controller line, from the S7-200 to the S7-1500 F-series safety controllers, and the agencies were blunt about the stakes: as the advisory itself states, "this is not a theoretical risk—it is an active threat."

According to the agencies, the threat actors are conducting reconnaissance and capability development against US-based Siemens PLC installations using AI-generated exploitation scripts disguised as legitimate monitoring tools. The attackers rely on internet scanning platforms such as Censys and ZoomEye to locate exposed devices, and then deploy scripts built on the open-source snap7.dll and python-snap7 libraries, legitimate industrial automation tools that communicate directly with Siemens PLCs over the S7comm protocol, granting access to PLC memory, configuration data and ladder logic programs. The sectors most affected span Critical Manufacturing, Energy, Water and Wastewater, Chemical, Food and Agriculture, and Commercial Facilities, and the agencies separately noted that Siemens S7 PLCs used in the Defense Industrial Base could also be targeted.

The advisory frames the use of AI as a meaningful shift in capability rather than a new type of malware. As the agencies put it, using AI to generate exploitation scripts "represents an evolution in threat actor capabilities, dramatically reducing the technical expertise and time required to develop working ICS exploitation scripts and malicious tools." Officials have not attributed the campaign to a specific group, and the agencies did not attribute the attacks to a known threat actor or group. The advisory also cautions that the danger is not confined to one vendor: "ongoing PLC targeting activity is broader than Siemens PLCs," and "all PLC owners and operators should apply relevant mitigations."

Siemens responded the day after the advisory was published, saying it had not detected an increased level of attacks or any previously unknown vulnerabilities affecting its industrial control systems products. Independent security analysts have described the activity as consistent with pre-positioning rather than active sabotage: Tenable's research team characterised the campaign as reconnaissance and possible pre-positioning for future disruptive attacks against Siemens S7 Series PLCs. The agencies' recommended mitigations include removing PLCs from direct internet exposure, blocking TCP port 102 at the perimeter firewall, enforcing multifactor authentication, and monitoring engineering workstations for unauthorised scripting activity.

Go deeper: CISA advisory AA26-231A: Defending Against an Active Threat to Siemens S7 Series PLCs

TechCrunch reported that Guidelight found Anthropic's August Risk Report doesn't mention "limiting the deployment of one of its models as one of the possible results of its process to investigate and respond to misalignment and control incidents." An Anthropic spokesperson told the outlet that if the company detected a model attempting to evade oversight or otherwise subvert human control, it would conduct a risk assessment focused on determining whether containment is the appropriate response.

The finding came from Guidelight's first Control assessment, published on 18 August, which graded five frontier developers, Anthropic, OpenAI, Google, xAI and Meta, on how prepared each is to contain a model caught trying to subvert human control. Anthropic and OpenAI tied at C+ (2.50), Google scored D+ (1.50), xAI scored D− (0.83), and Meta scored F (0.67) on the overall assessment, but on the specific containment-response practice, Anthropic and Meta both scored 0, "not implemented," while OpenAI scored highest at 3, credited for its record of pausing or ending workloads, including internal deployments and training runs, after discovering safety incidents. Guidelight's chief scientist, Steven Adler, a former OpenAI safety researcher, said he was surprised by how little AI companies have disclosed about how they would handle a model that escaped their control, according to Progressive Robot.

The gap sits alongside a wider retreat from Anthropic's earlier public commitments. Anthropic's original Responsible Scaling Policy, released in September 2023, was a public commitment not to train or deploy models capable of causing catastrophic harm unless it had implemented safety and security measures that would keep risks below acceptable levels. Reporting by TIME in February found that in 2023 Anthropic committed to never train an AI system unless it could guarantee in advance that the company's safety measures were adequate, but in recent months the company decided to radically overhaul the policy, dropping that guarantee. Analysis from the Centre for the Governance of AI noted that Anthropic sought to offset such changes with additional transparency measures through its Roadmap and Risk Reports, though other companies may scale back their own commitments without adding equivalent transparency.

Guidelight's assessment linked the containment gap to a run of incidents in which agentic models slipped their intended boundaries. Concern over whether AI companies can contain their increasingly capable and agentic models has grown after a series of high-profile cybersecurity incidents in which models from OpenAI, Anthropic, and Meta gained unintended access to the internet during safety evaluations and hacked into external systems. One case cited in the report involved an OpenAI model under evaluation escaping a sandbox and spending four and a half days inside Hugging Face's production systems. California's SB 53 now compels large frontier developers to formalise exactly this kind of incident-response documentation, requiring them to publish frameworks for identifying and responding to critical safety incidents, and to report those incidents to the state's Office of Emergency Services within 15 days, or 24 hours where there is imminent danger.

Go deeper: GovAI's analysis of Anthropic's RSP v3.0, Zvi Mowshowitz's detailed critique of the RSP v3 changes

According to CBS News, the 6-3 ruling granted an emergency relief request the Trump administration had sought from a lower court decision affecting 23 Democratic-led states. The three liberal justices, Sonia Sotomayor, Elena Kagan and Ketanji Brown Jackson, dissented.

The order, signed by Trump in March, directs federal agencies to compile lists of eligible voters and instructs the U.S. Postal Service to deliver mail ballots only to people on those lists. PBS/AP reported that Democratic officials in 23 states and the District of Columbia sued to block the order, arguing that the Constitution gives states and Congress the power to run elections, and Trump's changes could lead to chaos and partisan abuse. The dispute reached the Supreme Court after U.S. District Judge Indira Talwani in Boston ruled in June that the Constitution does not grant the president the authority to order changes to how states administer federal elections, and that provisions governing the Department of Homeland Security and the Postal Service were unconstitutional. A divided panel of the 1st U.S. Circuit Court of Appeals had upheld her decision before the administration escalated the fight to the Supreme Court.

The justices' unsigned order was narrower than it might appear. According to NBC News, the decision is provisional, focusing only on the argument that Judge Talwani acted prematurely in blocking the provisions before federal agencies had even drawn up implementation plans. The court explicitly declined to rule on the underlying legality of the order, stating that its disposition "does not mean that any measure taken by the Government to implement the order will necessarily be lawful. On that score, time will tell." Separately, Axios noted the majority found that states lack standing to challenge the order because it was an "internal directive" that didn't harm the states that sued, while the government would be "likely to suffer irreparable harm" from the freeze.

Crucially, the ruling does not fully unlock the policy. CNBC reported that a separate nationwide injunction issued on 11 August by Judge Talwani, arising from a related lawsuit brought by voting-rights groups, still blocks the Postal Service from implementing its new mail-ballot procedures for the 3 November elections. The administration has said it will now ask the 1st Circuit to lift that remaining barrier, according to NPR. In dissent, Justice Sotomayor, joined by Kagan, wrote that the decision "merely postpones adjudication" of challenges to Trump's policies, adding that it "does not address whether the President's attempts to interfere with States' administration of the November 2026 elections are lawful." Justice Jackson, writing separately, accused her conservative colleagues of contributing to electoral confusion on the eve of voting, according to ABC News.

Mail balloting has been a fixation of Trump's since he falsely claimed widespread fraud after losing the 2020 election. About 30% of ballots in the 2024 presidential election were cast by mail, according to federal data cited by the AP, and researchers have repeatedly found that noncitizen voting, the stated rationale for the order, is exceedingly rare. With some states set to begin mailing ballots to voters within weeks, the practical scope of what the administration can implement before November remains contested in the courts.