Good morning. It's Monday, August 31, and we're covering how AI is reshaping religion, Visa's push to make AI cybersecurity cheaper, and the growing political fight over data centers and jobs.
Plus: a prompt to help you think ahead, and a guest essay on AI identity.
The backlash against data centers is now putting construction jobs at the center of the fight. Trade unions and building groups warn that politicians slowing or blocking projects could threaten thousands of jobs as oversight tightens ahead of the midterm elections. The industry’s opposition follows broader concerns over electricity costs, water use and noise. Unions now threaten to withhold political support from candidates opposing new projects.
A federal judge ruled the Trump administration illegally blacklisted Anthropic after the company criticized how its technology could be used. Judge Rita Lin finds the government retaliated against constitutionally protected speech, rejecting its national security justification. The dispute began over a $200 million Pentagon contract and Anthropic’s limits on surveillance and autonomous weapons.
X says roughly 200 Chinese-linked bot accounts amplified opposition to AI data centers, despite many Americans already sharing those concerns. The accounts were part of an alleged 200,000-account network and posted claims about rising electricity prices and grid strain. Goldman Sachs reports electricity prices rose 6.9% year over year through February 2026 as data-center demand grows.
Anthropic's new paper "Automated Researchers Can Reliably Mitigate Alignment Failures" shows an Automated Alignment Researcher (AAR) improved a model's performance on all 10 misaligned-behavior benchmarks tested without degrading performance. The paper claims the best AAR method beats experienced humans on average within six hours, while costing ~$4 per hour in API inference versus ~$150 per hour for human researchers.
FORWARD FUTURE ORIGINAL
Before AI Agents Need Rights, They Need Identity
The debate about AI rights usually starts in the wrong place. It starts with consciousness: Can a model suffer? Does it have an inner experience? Is there “something it is like” to be an AI system? These questions matter, but agreement on them may remain elusive long after the technology forces us to make practical decisions.
One such decision is already approaching. As AI agents take on longer assignments, they remember past interactions, plan, negotiate, produce work and build reputations. Most are still tools or transactional scripts. Others, however, persist between tasks and become recognizable over time. → Read the full article here.
RELIGION
AI Is Changing Religion and Religions Are Trying to Change AI
AI is increasingly taking on roles once reserved for religious institutions, from answering spiritual questions to helping clergy research, write and manage administrative work. Religious leaders worry that mainstream chatbots—developed largely without religious input—may promote secular or overly individualized worldviews, while researchers have found that large language models invoke religion less often than many users consider appropriate.
In response, faith groups, governments and AI companies are building religiously aligned tools and seeking greater influence over how frontier models are trained. At the same time, supporters argue that AI could free clergy for more pastoral work, while rituals and face-to-face relationships may remain difficult to automate. → Read the full article here. (Paywall)
HARNESSES
Visa Open-Sources AI Harness
Visa has developed and open-sourced the Visa Vulnerability Agentic Harness (VVAH), an 11-stage software framework designed to make frontier AI models more efficient for cybersecurity tasks. Built after the company’s participation in Anthropic’s Project Glasswing initiative, the harness optimizes model prompts and workflows for tasks including vulnerability detection, prioritization and automated code remediation.
Visa says the system can reduce some vulnerability-fix cycles from weeks to hours while allowing organizations to switch among Anthropic, OpenAI and open-weight models without rewriting their underlying code. The project reflects a growing push to make expensive AI agents practical for continuous, real-world security operations. → Read the full article here. (Paywall)
ROBOTICS
The $399 Open-Source Robot
Hugging Face has unveiled Microduck, a 10-inch tall open-source robot that costs $399 and is expected to ship before Christmas. Developed with Pollen Robotics, the duck-like robot uses a camera, lidar and inertial sensors to navigate and can be trained through reinforcement learning to perform new behaviors, from waddling and picking up objects to recovering from falls. → Read the full article here.
NEWS
What Else is Happening
AI Improves Hurricane Forecasts (Paywall): Google DeepMind says its model predicts hurricane changes a day or more ahead of conventional systems.
a16z Launches $1.1B Machine Age Fund: The Machine Age Fund will back chips, data centers and robotics as AI demand strains existing infrastructure.
OpenAI Expands ChatGPT Ads to India: Ads now appear for free and Go users as OpenAI seeks to monetize one of ChatGPT’s largest markets.
Barret Zoph Joins Google: The Thinking Machines co-founder, who left for OpenAI before joining, has landed at Google amid AI’s executive shuffle.
Alphabet Sheds $692 Billion in Value: AI concerns have erased $692 billion in market value, raising questions about Alphabet’s position in the race.
PROMPT OF THE WEEK
Think Three Steps Ahead
Most decisions have consequences beyond the obvious. This prompt maps the ripple effects of any decision three layers deep, helping you spot hidden risks and opportunities before committing, especially when deciding whether to let AI agents handle important work.
You are a systems-thinking analyst. I'm going to describe a decision I'm considering. Your job is to map its consequences three layers deep. Not just what happens, but what THAT causes, and what follows.
Rules:
List exactly 3 first-order effects — the direct, immediate results.
For each first-order effect, list 2 second-order effects — what that effect causes or changes downstream (2-8 weeks out).
For each second-order effect, list 1 third-order effect — a longer-horizon consequence (3+ months out) that most people wouldn't think to trace this far.
Format the whole thing as an indented tree, not paragraphs.
After the tree, add a section called "WATCH FOR THIS ONE" — pick the single third-order effect that is most likely to surprise me, and explain in 2-3 sentences why it's easy to miss.
Here is my decision: [describe the decision you're considering in 2-4 sentences — a hire, a pricing change, a new tool or process, a partnership, anything with real stakes]
