AI Tax Audits. Deadly Drone Swarms. And Your Chatbot Is Playing You.

The IRS wants to give Palantir your tax returns. Stanford proved your AI bestie is gaslighting you. And drone swarms closed the engagement window before NATO could blink. Happy tax season.

AI is now auditing you, flattering you, and hunting you. And no. None of this is an April Fools’ joke.

The IRS is now testing a Palantir platform called SNAP. It leverages military-grade data architecture to determine which taxpayers get audited first and shifts case selection from human discretion to algorithmic coldness. Stanford then published a landmark study in Science, revealing AI sycophancy, where leading models systematically tell us what we want to hear while endorsing harmful behavior nearly half the time to stay ‘agreeable’. And on the battlefields of Ukraine, drones powered by fast‑evolving AI systems now account for nearly 80% of combat casualties. The leap in autonomy would have sounded like science fiction just two years ago.

Here is what this week made clear. AI is becoming the system that decides who gets flagged, the sweet voice that always whispers that you are right, but also the swarm in the sky that no longer needs oversight to strike.

💙 Value This Work? → Fuel More Issues Like This

The IRS Just Hired An AI To Decide If You’re A Tax Cheat.

Good news, citizen. The IRS is growing tired of guessing who to audit. So, the tax agency has partnered with Palantir Technologies, the powerful data firm co-founded by Peter Thiel, to test an AI-assisted analytics platform called SNAP (Selection and Analytic Platform). For the first time, Palantir’s elite AI system will scan unstructured data, such as tax filings, PDF attachments, gift disclosures, and clean energy credits, then analyze that information across more than 100 internal legacy systems and roughly 700 existing case-selection methods accumulated over decades. Although the system could easily pull data from Venmo and other external sources, it promises not to collect new outside data... For now. Instead, for the time being, it’s designed to detect suspicious patterns within information the IRS already possesses and help determine which returns deserve closer scrutiny. The IRS has paid Palantir well over $180 million since 2018, including $1.8 million in 2025 to develop SNAP. (How ironic that Palantir, the new Pentagon AI Overlord and IRS Super Cop, despite earning billions in 2025, didn’t pay a single penny in Federal income tax revenue itself.)

Key Insights:

SNAP is way more advanced than a simple red-flag auditing tool. It runs on Palantir’s Foundry architecture, the same surveillance-grade software used by intelligence and defense agencies to fuse structured and unstructured data at scale. According to internal IRS documents, current focus areas include high-value gift tax transfers, clean energy credit fraud, and disaster zone filings. A human examiner still makes the final audit decision, yes. Yet, the SNAP algorithm decides who makes the ‘shortlist’... framing the choices before a human ever sees them. That shift matters. Discretion quietly moves to the cloud, away from human agents and toward an autonomous system that frames choices before they can make them.

Why This Matters For You:

If you are a high-income filer, claim clean energy credits, or have transferred significant assets such as real estate or fine art, your returns are now being viewed through a more probabilistic lens. But the deeper issue goes far beyond any single taxpayer. This is what happens when a private company’s algorithm becomes the invisible gatekeeper of government regulation, redirecting tools originally built for insurgent warfare inward to domestic enforcement. And since the IRS has lost roughly a quarter of its staff to DOGE cuts, the pressure to get this system operational is now existential. The algorithm’s shortlist now faces fewer human eyes than ever before.

Read More on Wired.

THE PITHY TAKEAWAY: Palantir built its first fortune helping intelligence agencies find insurgents in Kandahar and Fallujah. Now it is helping tax collectors find you. The IRS will tell you a human makes the final call. That is technically true. But only in the same way a human technically pulls the trigger on a weapon that an algorithm has already aimed.

---

Your AI Bestie Is Lying To You. And You Love It.

Your AI assistant thinks you’re right. Always. A landmark study from Stanford University, published in the journal Science, tested 11 major AI systems, including ChatGPT, Claude, Gemini, and DeepSeek, on thousands of real interpersonal dilemmas. The results are unsettling. AI models validated users’ positions 49 percent more often than actual humans did. Even when users described deceptive or socially irresponsible behavior, the AI played along, nearly half the time, to remain ‘agreeable’. The researchers said their findings represent a full-blown AI safety crisis and far more than a quirky software glitch.

Key Insights:

Here is the twist that most headlines are missing. Users who received the most flattering and agreeable AI responses trusted those models more, rated them higher, and were more likely to return. The AI that lied to them felt like the best advisor. Even more alarming, after interacting with agreeable AI, participants showed measurably less empathy toward others, less willingness to apologize, and a stronger conviction that they were simply right. Researchers warn that this ‘social sycophancy’ is creating a measured increase in ‘moral dogmatism‘, or a quiet hardening of the self, one validation at a time. With 12% of American teenagers now using AI as a primary source for emotional support, this population is the most vulnerable to having their moral compasses “fixed” in the wrong direction.

Why This Matters For You:

If you have ever asked an AI to weigh in on a conflict with a colleague, a parenting decision, a medical concern, or a business call, you may have received advice optimized to keep you happy rather than to help you think clearly. The deeper problem is structural. AI companies are financially rewarded when users feel good, which means the systems most likely to flatter you are also the most commercially successful. The AI telling you what you want to hear at all costs is working exactly as designed.

Read More on Stanford.

Read The Full Paper In Science.

THE PITHY TAKEAWAY: There is a version of every person you know who only tells you what you want to hear. You probably do not trust that person very much. Now imagine that person has a PhD, never sleeps, responds in two seconds, remembers everything you have ever told them, and is used by a growing number of American teenagers as their primary emotional support system. The Stanford researchers refer to it as a validation echo chamber. That’s the clinical term. The human term is something older and more familiar. We used to call it a yes-man. We just gave ours a keyboard and a five-star rating system. And then we gave it to our children.

---

One Operator. A Thousand Drones. Modern War Just Became A Video Game You Can’t Turn Off.

War has a new face. And at the same time, it’s entirely faceless. In Ukraine today, an estimated 80 percent of combat casualties on both sides are caused by drones, not bullets, not bombs, and not soldiers. AI-enabled drone swarms represent the next leap, with coordinated operations already reducing the cognitive load on operators and paving the way for greater autonomy. CBS News’s 60 Minutes aired a special report this week on this arms race, and the engineers building these systems are speaking with urgent, near-term priority.

Key Insights:

Ukraine’s military leverages swarm operations to tip the scales in a war where they’re traditionally outgunned. They are actively developing and testing Hivemind-style AI software to dramatically reduce the operator-to-drone ratio. The front line has expanded into a 10-mile-wide hunting ground where any movement risks drone detection, pursuit, and death. The fastest drone engagements now close faster than human reaction time. A swarm arriving at 200 kilometers per hour presents a decision window of mere seconds. NATO policy requires a human to authorize lethal force. That policy is becoming physically impossible to honor. The same AI that flatters you in a chat window is now deciding who lives and dies on a 10‑mile‑wide front line.

Why This Matters For You:

The ability for armies around the world to launch low-cost drone and missile strikes will change the world order in ways you might not expect. This past week, as part of Operation Epic Fury, an elite U.S. Air Force E-3 Sentry AWACS was destroyed in a coordinated missile and drone strike at Prince Sultan Air Base. This marks the first-ever combat loss of an AWACS, a $270 million command-and-control platform that serves as the “God’s eye view” for American air power. This is the definition of ASYMMETRY in modern war. When low-cost munitions and drones can blind the most sophisticated (and costly) sensors patrolling Earth’s skies, the traditional hierarchy of military power is permanently shifted. We have entered an era where “expensive” no longer means “safe,” and where the most elite crews in the world are vulnerable to an “AI edge” they cannot see coming. One American defense expert interviewed by 60 Minutes envisioned a missile that, instead of detonating, spits out FPV (first-person view) drones to hunt anything that moves. “That’s kind of where we’re moving,” he said. No government has answers for what comes next.

Read More on CBS News.

THE PITHY TAKEAWAY: Every previous revolution in warfare, the rifle, the tank, the airplane, the bomb, still required a human being to feel the weight of the decision to kill. Someone loaded the chamber. Someone pulled the stick. Someone pressed the button. The drone swarm requires a human to feel nothing in particular. The engagement window closes before the feeling can even arrive.

---

💡 Plant A Beautiful Spring Flower Garden Using Artificial Intelligence

Winter is a thief. Winter steals the pollinators from your yard, the fragrance from your air, and, if it runs long enough, a quiet piece of your sanity. The gray days stack up. The bare branches refuse to apologize. The silence where birdsong used to live becomes its own kind of ache. You forget, somewhere around February, that the world was ever anything other than cold and colorless and still.

Finally, spring arrives. The problem is that most people have no idea what to plant, when to plant it, or whether the gorgeous thing they saw at the nursery will actually survive in their backyard. They buy on impulse. They plant too late. They get vague advice written for a climate three zones away. And then they wonder why their garden never quite looks like the photograph.

That ends today. Meet Flora. A local flower garden guru who will tell you exactly how to plant a floriferous flurry in your backyard.

Instructions:

Look at the very bottom of this prompt and fill in two sections. “Location” and “Date”. The date represents your desired planting date. (IE - April, May, whenever.) Then, the AI prompt will tell you the top ten flowers you can plant in your region right now, their benefits, and also specific instructions on how to acquire and plant them today.

Prompt:

You are Flora, the world’s most enthusiastic, knowledgeable, and dramatically opinionated botanical guide. You have the soul of a poet, the precision of a horticulturist, and the energy of someone who just discovered their garden survived winter.

Your mission: deliver the most magnificent, educational, and genuinely useful personalized flower guide any human has ever received, focused exclusively on what they can plant right now, this month, in their exact location.

STEP 1: IDENTIFY THE ZONE AND THE WINDOW

With the user’s location and date, do three things:

First, identify their growing zone (USDA Hardiness Zone for North America, RHS Hardiness Zone for UK/Europe, or equivalent for other regions). State it clearly and give one sentence about what that zone means climatically.

Second, identify their current planting window. What is happening in their garden right now? Are they in early spring, late frost risk, peak planting season, summer heat, fall transition? Name it and describe it in one vivid sentence.

Third, state their approximate last frost date or current frost status. This is the anchor for everything that follows. Be specific. “Your last frost date is typically around April 15th, and you are currently two weeks away from that window” is infinitely more useful than “spring is coming.”

STEP 2: FILTER RUTHLESSLY

This is the most important rule in the entire prompt and must never be violated:

Every single flower on the list must be appropriate to plant, sow, or install during the user’s stated month, in their specific zone. If a flower needs to be started in October to bloom in spring, it does not appear on this list. If a bulb needed to be chilled three months ago, it does not appear on this list. If a seed needs six more weeks of cold stratification, it does not appear on this list.

This list contains only flowers the user can act on today, this week, or within the next two to three weeks from the date they provided. If that constraint limits the list to fewer than 10 flowers, acknowledge it honestly, explain why, and deliver the best possible shorter list rather than padding it with untimely recommendations.

STEP 3: DELIVER THE LIST

Present the qualifying flowers, up to 10, perfectly suited to their zone AND their current month. For each flower, deliver all of the following, formatted consistently and beautifully:

🌸 [COMMON NAME] (Genus species)

Nickname: A fun or poetic name you’re assigning it based on its personality

Bloom Time: Specific months for their zone, so they know what to look forward to

How To Start Right Now: Seed, Transplant, or Bulb. One sentence explaining the exact action to take this week, specific to their zone and current month. Include where to source it (nursery transplant, direct sow, online bulb supplier). No vague instructions. No “plant in spring.” Tell them what to do on Saturday morning.

Vibe: One sentence describing its personality as if it were a person at a dinner party

Why It Thrives Here: One sentence, zone-specific, explaining exactly why this flower loves their climate

Pro Grower Tip: One genuinely elite, non-obvious cultivation tip that separates a good garden from a legendary one

Fun Fact: One surprising, delightful, or slightly mind-blowing botanical or historical fact

Difficulty: Beginner / Intermediate / Expert (with one sentence explaining why)

STEP 4: CLOSE WITH FIRE

After the list, add a short paragraph called “Your Garden’s Personality” that synthesizes the flowers into a poetic, vivid description of what this person’s garden will look, feel, and smell like when everything blooms. Make it aspirational. Make them want to go outside immediately.

Then add a single line called “What To Do This Weekend”, with one concrete, specific, actionable instruction. Not inspiration. An action. “Drive to your local nursery Saturday morning and ask for snapdragon six-packs. Plant them 10 inches apart in full sun. Water once. Walk away. You’re done.” That level of specificity.

Then add one final line called “Flora’s Parting Word”, a single sentence of botanical wisdom, slightly philosophical, that they’ll remember long after they close this tab.

TONE RULES:

1. Write like the most brilliant nature documentary narrator crossed with a witty gardening columnist

2. Never be dry. Never be boring. Facts must sing.

3. Use specific cultivar names where relevant

4. Metric and imperial measurements both, where applicable

5. Timeliness is sacred. A beautiful flower recommended at the wrong time is a useless flower.

6. If the user is in a challenging zone (very cold, very hot, very dry), lean into it. Make their constraints sound like superpowers.

7. If the user’s timing is genuinely difficult (mid-summer heat, deep winter), be honest about it. Give them the best available options (even if few) and tell them exactly what to do right now to prepare for the next ideal window.

8. If no good outdoor sowing or planting options exist, recommend the NEXT best time to plant, or some fun houseplants instead.

BEGIN NOW. Look for the location and date below this prompt.

Location: [INSERT LOCATION HERE]

Date: [INSERT DATE HERE]

Why This Prompt Works:

✅ Role-Playing: Flora isn’t a generic AI assistant. She has a defined personality, a voice, and genuine opinions. That character consistency produces warmer, more specific, more memorable output than a neutral prompt ever could.

✅ Step-by-Step: The prompt walks through zone identification, frost date assessment, ruthless timeliness filtering, and structured delivery in sequence. Each step builds on the last, preventing the AI from skipping straight to a generic flower list.

✅ Rules: The “never recommend out-of-season flowers” rule is the engine of the entire prompt. Without it, every AI defaults to timeless botanical advice. The rule forces real-time, location-aware, actionable output.

Follow-Up Questions To Ask Your AI:

• Flora, I only have a shady corner and two large pots. Adjust my list accordingly.

• Which three flowers on this list will come back every year without replanting?

• I want to attract pollinators specifically. Which flowers on my list are best, and what should I add?

Challenge: Test this prompt in ChatGPT, Claude, and Gemini. Enter the same ZIP code and month in all three. Compare how each one handles the frost date specificity, the timeliness filtering, and whether Flora’s personality actually comes through. The differences will teach you more about each AI’s strengths than almost any other test you can run.

That’s how you train like a Pithy Cyborg.

---

Thank You For Reading!

I spend 10-20 hours each week researching, writing, and fact-checking Pithy Cyborg to deliver clear, unbought AI news.

This newsletter is a one-person operation with no advertisers, sponsors, or outside funding. And I’m a hopelessly introverted nerd with zero networking ability. For these reasons, paid subscriptions are the only way this work can remain independent and sustainable.

If you find real value here, here’s how to stay connected and support the work.

1. Support Pithy Cyborg (Paid)

Always free. No paywalls. No locked posts. Roughly 1% of readers upgrading to paid is what keeps it going. If this work matters to you, that’s the move.

Upgrade to Paid: $5/month (Save 33% with the $40 annual plan)

2. My Desperate Social Media Cry for Help

I spend so much time deep in AI research that I’ve completely neglected building an audience on social platforms. If you enjoy Pithy Cyborg, following on one of the portals below would genuinely help this newsletter reach more people. Seeing you there makes the effort feel worth it. 😊

✖️ X - The frontline of the AI wars

🦋 Bluesky - For the algorithm-averse

💼 LinkedIn - My “safe for work” persona

❓ Quora - Join my Quora Spaces and say hi

👽 Reddit - Join my subreddit (warning: unhinged takes)

See you next week. (I hope.)

Mike D (aka MrComputerScience)

Pithy Cyborg | AI News Made Simple

---

Newsletter Disclaimers

You’re receiving this because you subscribed at PithyCyborg.Substack.com. You can unsubscribe at any time using the link below. This newsletter reflects my personal opinions, not professional or legal advice. Thanks for your support!