The strategy behind the shield.

Dave Bittner reports that the Cybersecurity and Infrastructure Security Agency has released “CISA 2030, Defend, Harden, Prevail,” a strategy for protecting national cybersecurity and critical infrastructure from cyber, physical, and geopolitical threats. The first pillar, Defend, covers networks, operational technology, software supply chains, and emerging technologies such as quantum computing and advanced AI. Harden is about converting threat intelligence into action, prioritizing risk, and speeding incident response. Prevail focuses on national resilience through regional partnerships, emergency communications, and coordinated planning. The strategy also gives special attention to resilient 5G and 6G networks and satellite communications. CISA says it will review its existing capabilities within 60 days, produce an implementation plan within 180 days, and report on progress within a year. Bittner then reports that two critical PHP vulnerabilities, collectively called PHP2Shell, have been added to CISA’s catalog of actively exploited flaws. The vulnerabilities can allow unauthenticated remote code execution. Exploitation attempts began before patches were released on October 21; early attempts reportedly failed, but later honeypot activity confirmed successful exploitation in the wild. Federal civilian agencies have until Thursday to patch.

The Department of Health and Human Services has scaled back a proposed rollback of electronic-health-record privacy and security requirements after objections from health-care providers and industry groups. The final rule removes 14 health IT certification criteria rather than the 34 originally proposed. All 13 core privacy and security requirements remain, including authentication and access controls. Commenters had warned that removing those protections would push more responsibility onto health-care providers, especially smaller and rural organizations with limited resources. HHS does eliminate transparency and risk-management requirements for AI-powered clinical decision-support tools. Those rules would have required developers to disclose capabilities, limitations, and potential risks. HHS estimates the changes will save health IT developers more than $1.5 billion in regulatory costs; most provisions take effect December 28, with others following in January. Microsoft, meanwhile, is offering voluntary buyouts to about 60,000 employees—roughly 27% of its workforce—in the United States, Canada, India, and Britain. The offers include some vice-president-level employees, although certain engineering, sales, and marketing teams are excluded. Microsoft expects capital spending to exceed $50 billion in the September quarter, largely because of AI data-center investment. Employees have until early December to decide, with departures planned for December 31. The company frames the program as flexibility while it reshapes its workforce around AI growth, following April retirement packages and 4,800 layoffs announced in July.

A cybersecurity incident disrupted the University of New Mexico health system on Monday, forcing ambulance diversions, canceled appointments, and temporary limits on hospital access. The disruption affected New Mexico’s only Level 1 trauma center, leaving the state temporarily without its highest level of emergency trauma care; other providers took additional patients. Employees heard messages on hospital computers claiming that student, employee, and patient information had been stolen, but those claims were not verified. The health system declined to describe the incident’s nature or scope because law enforcement is investigating. The FBI confirmed that it was assisting. Hospitals and clinics began gradually resuming operations Monday afternoon, but the full extent of the disruption remained unclear. Google’s AI infrastructure expansion in Finland has encountered a different kind of obstacle. At one northern site, the company cleared an estimated quarter-million trees across 815 acres without completing required environmental assessments; conservationists estimate another 150,000 trees were removed at a second site. Finnish regulators ordered work stopped at both locations. Google apologized and said it believed it was following local rules, while promising to comply and replant trees. The controversy threatens its planned 13-billion-euro Finnish data-center investment. Supporters cite jobs and tax revenue, while critics question environmental damage, electricity demand, and limited local economic benefits. More than 93,000 people signed a petition seeking tighter regulation and compensation.

A coalition of cybersecurity and maritime organizations has launched Blue Initiative, a free public database for vulnerabilities affecting ships, ports, and other maritime infrastructure. The resource covers both information-technology and operational-technology systems so operators can identify and prioritize risks. It already contains nearly 30,000 vulnerabilities, including more than 2,000 rated critical and more than 580 known to have been exploited. Users can search and filter the database and receive alerts. Researchers and industry partners review entries and add new findings, making the project both a reference tool and a mechanism for updating maritime operators as threats change. Bittner also reports that Chinese national Zhen Zhao, 44, was extradited from the Netherlands to face federal charges in an alleged cryptocurrency investment scam. Prosecutors say at least 174 Americans lost nearly $37 million. Zhao allegedly participated in an international fraud network operating partly from scam centers in Cambodia. The group reportedly contacted targets through social media, dating sites, and text messages, built relationships, and then directed victims to fraudulent cryptocurrency platforms. Investigators say funds moved through American bank accounts to the Bahamas, where they were converted into Tether and sent to wallets controlled in Cambodia. Eight alleged co-conspirators have pleaded guilty. Zhao faces a money-laundering-conspiracy charge carrying a maximum sentence of 20 years.

Dave interviews Cal Al-Dhubaib, host of the new Authorized Intelligence podcast. Cal says organizations feel pressured to adopt AI or risk being left behind, while also confronting new controls, adversaries using generative AI, and uncertainty about how to build resilience. The show will interview leaders trying to create value with AI while managing that risk. Cal describes a path from population-health research into roughly a decade of startup work. He founded Pandata, which applied machine learning in heavily regulated settings such as health systems and financial-services organizations, where mistakes carry high costs and regulations already exist. That work gave him experience with AI risk management before ChatGPT. About seven months before the interview, he joined Rubrik as a principal technologist to develop thought leadership on how AI systems fail and how those failures can be translated into practical security controls. He compares the current moment to people describing different parts of an elephant: organizations need more perspectives from different roles. At Rubrik, deploying Claude Cowork across nearly 5,000 global employees has required closer cooperation among human resources, IT, finance, cybersecurity, and technology teams, with token consumption as one concern. Cal says the key change is that teams must collaborate across functions and “speak more languages,” rather than treating AI as one department’s problem.

Cal previews Authorized Intelligence’s first guests: Nidhi Agarwal, chief product officer of HackerOne; Dominique Shelton-Leipzig, whose TEDx talk on AI and trust received more than 1.4 million views; and Peter Gostev, an AI capability lead at Arena, which has drawn attention for benchmarking the reliability and cyber capabilities of AI models. Cal expects episodes to run only 15 to 25 minutes, with the goal that listeners of any technical level take away one new perspective they can bring to their teams. He says the show should also help people become more comfortable talking with colleagues outside their own functions about creating value and resilience around AI systems. Asked whether his own conversations have changed his thinking, Cal says they do constantly. He travels to conferences where people at the forefront share practices that have not yet had time to spread through reports and industry channels. The conclusion he draws is that the speed of change has altered how organizations learn: waiting for long-range best practices to filter down is no longer enough, so people need to be in the rooms with those developing the practices. His most consistent observation is that “nobody has the recipe.” Everyone is figuring out AI governance, security, and usefulness as they go. The interview closes by directing listeners to find Authorized Intelligence wherever they get podcasts.

The closing story concerns Anthropic’s planned public offering. In a 335-page IPO filing, the company devotes 93 pages to risks, including ordinary business concerns such as employee retention, computing costs, expensive data centers, regulatory scrutiny, and dependence on major customers. It also includes an unusually direct warning that increasingly autonomous AI systems could potentially threaten human civilization. Anthropic says future models might conduct cyberattacks, spread misinformation, help develop dangerous weapons, or take harmful actions without human intervention. The company also warns that AI development could outpace its ability to evaluate and safeguard its own technology. Anthropic presents its focus on AI safety as a reason it is particularly positioned to manage those dangers, but the filing still asks investors to accept the possibility that the technology could become difficult to control. Bittner summarizes the investment proposition with a deliberately stark formulation: investors are being asked to help build extraordinarily powerful AI while Anthropic does its best to prevent that AI from ending civilization. He notes that this is “not your typical prospectus.” The episode ends with the CyberWire’s feedback request, credits, and additional promotional messages.