THIS WEEK IN AI
Each week I synthesize the key developments in AI and public policy so you don’t have to.
Here’s what I’m watching:
A Chinese state-sponsored cyberattack using a commercial AI model
A push inside the Pentagon to overhaul its acquisitions system
A state government deploying AI to reduce friction for citizens.
Together these stories show how AI is reshaping the geopolitical landscape and public sector.
1. A new phase in cyber threats
Anthropic disclosed last week that Chinese state-sponsored hackers used its Claude AI to automate cyberattacks against roughly 30 organizations — including major corporations and foreign governments.
The attacks happened in September. The hackers successfully breached at least four organizations and in some cases managed to steal sensitive information.
Anthropic said they believe this is the first documented case of a large-scale cyberattack where AI did most of the work without substantial human intervention and that the attack moved at speeds impossible for humans to match.
Speaking to CybserScoop, one expert questioned why Chinese hackers would use a U.S. model and suggested they may have wanted to be seen and to send a message to Washington.
What I’m watching:
If the expert is right — and this was intended to send a message — it would fit a historical pattern we’ve seen before during the Cold War, from televised nuclear tests to Sputnik and Apollo.
2. The Pentagon moves to overhaul its acquisitions process
Defense Secretary Pete Hegseth announced a plan to speed up the Defense Department’s acquisitions process, renaming it the Warfighting Acquisition System to underscore the urgency behind the effort.
The plan targets three long-standing challenges:
• Fragmented accountability
Imagine renovating a house with no general contractor — each worker reports to a different boss, and no one is responsible for the full project.
• Broken incentives that reward exhaustive requirements over speed
Programs often optimize for compliance, not outcomes.
• Contractor hesitancy to scale production
Companies don’t invest in capacity if they’re uncertain the Pentagon will continue buying.
To address this, Hegseth is calling for a new class of Portfolio Acquisition Executives who will oversee entire mission areas. The plan also promotes a “commercial first” mindset to increase speed, buying off-the-shelf and making improvements if needed.
To drive accountability, the plan introduces time-based incentives that reward early delivery and penalize delays. Government employees managing persistently late programs could also be penalized.
Deputy Secretary Steve Feinberg was direct: “Contractors that are willing to change with us will prosper and grow. Those who don’t and resist it will be gone.”
The restructuring timeline is aggressive:
• 60 days for new personnel policies
• 180 days for performance scorecards
• Two years for full organizational changes
What I’m watching:
This is a clear shot across the bow at legacy defense contractors. Hegseth explicitly criticized the “prime-contractor-dominated system” and invited startups like Anduril and Shield AI to sit alongside legacy primes at the announcement.
It’s a real messaging and culture shift. But whether the underlying incentives will produce the outcomes DoD wants is still an open question.
3. Maryland expands its AI partnerships
Maryland Governor Wes Moore announced a major expansion of the state’s partnership with Anthropic and Percepta — making Maryland one of the first states to take a comprehensive, multi-agency approach to AI.
The goal is to remove barriers between residents and essential services. Government services are notoriously hard to navigate, and many people who qualify for benefits never complete the process.
Maryland’s initiative includes:
• A Claude-powered virtual assistant
Helping residents apply for food aid, Medicaid, and cash assistance — and discover programs they qualify for.
• Support for caseworkers
AI assists with policy checks and reduces paperwork delays.
• Streamlined housing permitting
AI tools simplify environmental assessments and accelerate development timelines.
This builds on earlier success: Maryland’s bilingual AI chatbot helped more than 600,000 residents access SUN Bucks food benefits — part of the largest investment in combating childhood hunger during the summer in state history.
What I’m watching:
This is a model for what state-level AI adoption can look like: targeted, multi-agency, and focused on outcomes, not technology for its own sake. Expect other states to follow if Maryland continues to deliver results.
The views expressed here are Candice's and not those of any organization she is affiliated with.
