AI in Marketing

Ezra Klein, in his New York Times column, argues that the speculative questions about AI have become present-tense realities. The immediate case in point: the Pentagon's falling out with Anthropic.

Anthropic signed a defense contract last July with two conditions: no mass surveillance and no lethal autonomous weapons. The Pentagon found these restrictions unacceptable, particularly a clause barring the use of Claude to analyze bulk commercial data. Rather than simply ending the contract, Defense Secretary Pete Hegseth designated Anthropic a "supply chain risk," which would effectively prohibit any Pentagon contractor from doing business with the company. Even former Trump aides called it corporate destruction by government fiat.

Klein tries to take both sides seriously. AI models aren't mechanistic tools; they make contextual judgments guided by internal principles, and Anthropic's constitution instructs Claude to prioritize ethics even over the company's own directives. The Pentagon's demand for an "any lawful use" standard sounds reasonable until you consider that this administration regularly tests the boundaries of the law, and that existing laws weren't written with AI capabilities in mind. On the other hand, elected officials are more accountable than tech CEOs, and a private company setting red lines on military operations raises its own problems.

The Pentagon subsequently signed with OpenAI, which claims it secured the same red lines Anthropic insisted on. Klein suspects the real difference was political alignment, not contract language. He ends by noting that Congress has done almost nothing to establish legal frameworks for government AI use, held back by a combination of ignorance and fear of the industry's political spending.

Next.

I uninstalled OpenClaw after seven days. The experiment cost me about $150 in API fees and more hours than I want to count.

The first problem was safety. OpenClaw needs deep system access to be useful, so I ran it in a virtual machine. But sandboxing it for safety while giving it enough reach into my email, calendar, and files to do anything worthwhile was a contradiction I never resolved. I spent more time configuring the tool than using it.

Then there was cost. The Anthropic API burned through credits fast. I switched to DeepSeek, which was cheaper but painfully slow.

What really killed it was that Claude kept getting better in parallel. Scheduled tasks, workspace integration, Claude Code. I was already using Claude alongside OpenClaw to check its work or handle what it couldn't, and at some point the agent became overhead rather than help. The things I wanted from an autonomous assistant were increasingly just available in the tools I was already paying for.

Moving on.

A study led by Bournemouth University reveals that 41% of UK adults are open to using AI for mental health support, with 61% globally sharing this sentiment. The research, involving 31,000 participants across 35 countries, also indicates that a quarter of UK adults would consider using AI for teaching their children, and 45% globally would trust AI as their doctor. The findings highlight a significant willingness to integrate AI into personal and professional roles, though concerns remain about the quality and safety of AI advice. Dr. Ala Yankouskaya, the study's lead, notes that while AI can offer immediate support, it lacks the precision of human professionals. The mental health charity Mind has launched a commission to explore AI's role in mental health, emphasizing the need for responsible development and safeguards. The study, published in AI and Society, underscores the potential and challenges of AI in healthcare and education.

Meanwhile.

Caitlin Kalinowski, a senior member of OpenAI's robotics team, resigned due to concerns over the company's partnership with the U.S. Department of Defense. Announced on social media, her departure was based on principles regarding the use of AI in national security. Kalinowski highlighted insufficient policy guardrails, particularly around surveillance and autonomous weapons, as reasons for her decision. She emphasized her respect for OpenAI's leadership, including CEO Sam Altman, and expressed pride in her team's work. OpenAI stated that the agreement with the Pentagon aims to responsibly integrate AI into national security, explicitly rejecting domestic surveillance and autonomous weapons. The resignation occurs amid increased competition among AI developers to supply technology to the U.S. government, with tensions rising between OpenAI and Anthropic. Anthropic's CEO opposed using AI for mass surveillance or autonomous weapons, clashing with defense officials who seek flexibility in AI deployment. Kalinowski intends to continue working in responsible physical AI after a brief hiatus.

On a different note.

Jack Dorsey, cofounder and CEO of Block, recently laid off 40% of his workforce, citing the need to restructure the company around artificial intelligence. In an interview with WIRED, Dorsey explained that advances in AI tools necessitate a leaner, more agile company structure. He emphasized the importance of staying ahead of technological trends and expressed a desire to proactively adapt rather than react. Dorsey dismissed claims of overhiring, stating that Block's gross profit per employee was competitive with industry peers. He acknowledged employee dissatisfaction with the layoffs but maintained that the severance packages were generous. Dorsey envisions a future where companies operate with minimal management hierarchy, integrating AI as a core component. He believes this shift will allow customers to create personalized products and experiences. Dorsey also discussed Block's focus on Bitcoin, viewing it as an open protocol for money transmission. He expressed disappointment with the centralization of the digital world but remains optimistic about decentralization. Regarding Elon Musk's management of Twitter, Dorsey appreciates the company's privatization but criticizes some algorithmic decisions and the fragmentation of conversations. He disapproves of mocking users and values community-driven initiatives.

Finally.

Microsoft Threat Intelligence reports that North Korean threat groups are increasingly using artificial intelligence to enhance their schemes of infiltrating global companies by posing as remote technical workers. The groups, identified as Coral Sleet, Sapphire Sleet, and Jasper Sleet, employ AI to create convincing digital personas, conduct research on job markets, and align fake profiles with targeted roles. AI tools are also used to craft sophisticated social engineering tactics, including media creation for impersonations and real-time voice modulation. These technologies allow operatives to maintain long-term employment and evade detection by generating professional responses and technical solutions. While generative AI is currently predominant, a shift towards agentic AI is anticipated, potentially enabling more advanced and autonomous threat activities, although large-scale use has not yet been observed.