The only thing ringing in my head right now is the Iranian Lego videos.

A group calling itself Explosive News, known in Persian as Akhbar Enfejari, describes itself as an Iranian Lego animation team. A note on language: Farsi is simply the Persian word for the Persian language, the way Deutsch is the German word for German. The Academy of Persian Language and Literature maintains that Persian is the appropriate designation in English.¹

From the first day of the US-Israel war on Iran, Explosive News flooded global social media with AI-generated animations targeting American and Israeli audiences. One video opens with Lego versions of Donald Trump and Benjamin Netanyahu flanked by the devil, examining an album labelled the “Epstein file,” before Trump hits a red button launching a missile. Another features an Iranian military commander rapping over footage of Lego caskets draped in American flags, ending with falling blocks forming the letters L-O-S-E-R. The GB News TikTok post of one clip pulled 1.2 million likes.

Iran’s state broadcaster, Channel 3, aired a separate Lego-style animation on June 24, 2025, recapping the earlier Twelve-Day War, depicting missile strikes, downed aircraft, and crowds in Lebanon, Iraq, Yemen, and Palestinian territories celebrating. The Revayat-e Fath Institute was linked to several productions. The group denied state ties, describing themselves to the Associated Press as students producing content independently, though analysts told Fortune the bandwidth needed to generate and upload this content from inside Iran’s blacked-out internet indicated unofficial or official government cooperation.

Say what you want about the messaging. The communication strategy was genius. And yes, Jesus still loves us.

What Made Iran’s Content Work

The Explosive News team did not make random content. They studied the American psyche for weeks or used their own knowledge source before producing those videos. They knew the Epstein files were an open wound. They knew Lego felt safe. They knew rap was the global language of resistance.

Three things made it travel. The format disarmed the audience. The cultural references required no explanation. The content required no agreement to share. You did not need to support Iran to forward the video.

Snopes verified the videos as authentic Iranian propaganda. Radio Free Europe noted the United States was focusing its memes on domestic audiences while Iran was engineering content for global reach. The US and Iran were not fighting the same information war.

That is the communication lesson. Not the politics. The craft.

The War: Both Sides

It is essential to state what this conflict involved on both sides. It is equally essential to note that independent verification from inside Iran was severely constrained throughout. Iranian authorities first cut internet access on January 8, 2026, to suppress coverage of nationwide protests, and connectivity dropped to roughly 4% of normal levels following the February 28 attacks. US and Israeli cyber operations also contributed, targeting government news sites and IRGC communications infrastructure.

What the US and Israel did:

The United States and Israel launched joint airstrikes on Iran on February 28, 2026, during what Iranian officials described as active nuclear negotiations. The operation, codenamed Epic Fury by the US, killed Supreme Leader Ayatollah Ali Khamenei on the first day, confirmed by both Iranian state media IRNA and Israeli Defence Minister Israel Katz. Additional officials killed in the opening strikes included Ali Shamkhani, secretary of the Iranian Defence Council, and Mohammad Pakpour, commander of the Islamic Revolutionary Guard Corps, both confirmed by IRNA.

By April 7, 2026, Human Rights Activists in Iran (HRANA), a US-based NGO, documented 3,636 deaths in Iran from strikes, comprising 1,701 civilians, 1,221 military personnel, and 714 unclassified.² Iran’s Ministry of Health reported separately that 220 children under 18 were killed and 1,959 injured, with 18 of the dead under five years old. Iran reported damage to at least 120 historical heritage sites by late March.

On March 9, 2026, an Israeli airstrike on Tehran’s Resalat neighbourhood destroyed a Basij-affiliated building and three adjacent residential buildings, killing between 40 and 50 people. BBC analysis confirmed Israel used Mark 82 bombs. BBC Verify also confirmed a separate strike on residential buildings and a sports hall in Lamerd, killing 21 people including four children, was carried out using a US Precision Strike Missile.

The school in Minab:

The most contested event of the war was the strike on the Shajareh Tayyebeh girls’ elementary school in Minab on February 28, 2026. IRNA, Iran’s state news agency, reported 168 children and 14 teachers killed. UNESCO and UNICEF both issued statements of alarm. Two sources briefed on preliminary US military investigation findings told CNN the strike was deemed accidental, likely due to outdated targeting information about a nearby naval base. The Washington Post reported the school had been on a US target list. Israel denied involvement. No independent agency reached the site during active hostilities.³

Iran’s Atomic Energy Organization condemned a separate US-Israeli strike on the Shahid Rezayee Nejad yellowcake production facility in Ardakan, Yazd Province, calling it a “clear violation of the immunity of peaceful nuclear facilities.” Tasnim News Agency, affiliated with the IRGC, reported that a strike on Sharif University of Technology in Tehran destroyed its computing centre and GPU facility. Iran subsequently threatened to strike the OpenAI-Oracle-NVIDIA Stargate AI centre in Abu Dhabi in response.

What Iran did:

Iran launched hundreds of drones and ballistic missiles at Israel and at US military bases across Bahrain, Jordan, Kuwait, Qatar, Saudi Arabia, and the UAE. A drone struck Britain’s Akrotiri military base on Cyprus. The UAE intercepted 26 drones but nine fell inside its territory. Drones struck Oman’s Salalah port fuel tanks. The US Embassy in Kuwait was hit and closed indefinitely.

Inside Israel, the largest single death toll from an Iranian strike was in a residential neighbourhood in Beit Shemesh, where nine civilians were killed on March 1. A missile hit Haifa on April 5, killing at least two. Cluster bomb warheads struck Ramat Gan, killing two elderly residents outside their safe room. Debris from an Iranian missile fell on the Old City of Jerusalem, near the Al-Aqsa Mosque and the Church of the Holy Sepulchre.

The US military confirmed 13 service member deaths from Iranian attacks across the region, with 140 wounded by day 12. Six US aircrew died when a refuelling aircraft crashed in western Iraq.

Iran closed the Strait of Hormuz, driving oil prices above $170 per barrel and triggering European market selloffs. A 2 weeks ceasefire took effect April 8, 2026, brokered through Pakistan, after 40 days of conflict.

UN Secretary-General António Guterres condemned the operation at an emergency Security Council session, saying it had “squandered a chance for diplomacy.” UN Special Rapporteur Ben Saul described the strikes as “not lawful self-defence.”

Both sides caused civilian harm. Both sides contested each other’s narratives. Independent verification from inside Iran was severely limited throughout.

The AI War Layer

The Real Blocks Being Broken: The Parallel AI War

Underneath the bright colors, the actual war was grey, cold, and devastatingly efficient. This was the world’s first true “AI War,” where algorithms moved faster than diplomats.

The United States deployed the Maven Smart System, built by Palantir and powered by Anthropic’s Claude model, to provide real-time targeting and weapon recommendations.⁴ In parallel, Israel used AI systems known as The Gospel and Lavender, developed by Unit 8200. These systems, first scaled during the Gaza conflict—where they were reportedly programmed to accept up to 100 civilian casualties per strike on a single combatant⁶ were now fully active in Iran and Lebanon. In the first 12 hours of the offensive, nearly 900 targets were struck; over 2,000 in the first week.⁴

But the world was shocked not just by the technology but by Iran’s structural resilience. When Operation “Epic Fury” killed the Supreme Leader and dozens of top officials on the first day, Western analysts expected a total collapse. Instead, Iran’s “Mosaic Defense” doctrine kicked in. The military had been restructured into 31 independent provincial commands. This decentralized, devolved system meant that when the “head” was removed, the “tentacles” kept striking. Each command acted as its own mini-state with independent authority to launch drones and missiles.

Iran fought the information layer with equal technological sophistication. Groups like MuddyWater and CyberAv3ngers deployed AI-assisted polymorphic malware. They used Large Language Models (LLMs) to rapidly iterate on code, creating Rust-based backdoors like CHAR and GhostBackDoor that evaded traditional security by constantly changing their signatures. They weren’t just guessing passwords; they used AI to generate sophisticated Shodan and Google Dork queries at scale, scanning for internet-exposed industrial control systems (ICS). This leverage enabled the compromise of public surveillance networks and critical infrastructure, turning the very “eyes” of their adversaries against them in real-time.

Furthermore, Iran integrated Chinese radar systems and satellite data (via the BeiDou constellation) to track stealth aircraft, assisted by engineers from CETC institutes. Even more surprising was the use of TikTok for real-time intelligence. While the public saw Lego videos, the IRGC used recommendation algorithms to crowdsource battle damage assessments and track troop movements through “accidental” civilian uploads. It was a war where the “eyes” were everywhere.

 

600 by 600
Claude.ai

The Anthropic Situation

In July 2025, Anthropic signed a $200 million contract with the Pentagon, making Claude the first frontier AI model approved for use on classified US military networks.⁷ The contract included two restrictions: Claude would not be used for fully autonomous weapons, and it would not be used for mass surveillance of US citizens.

Negotiations over these terms broke down. On February 27, 2026, Trump ordered all federal agencies to immediately cease using Anthropic’s technology. Defense Secretary Pete Hegseth declared Anthropic a supply chain risk, making Anthropic the first American company ever given that label.

A California federal judge, US District Judge Rita Lin, granted Anthropic a preliminary injunction on March 26, 2026, ruling the designation appeared to be retaliation for First Amendment-protected speech. A DC appeals court denied Anthropic’s separate request, ruling the balance of harm favored the government during an active military conflict. OpenAI signed its own deal with the Pentagon hours after Anthropic was blacklisted. Anthropic’s lawsuit continues.

On the Cybersecurity Side

Anthropic launched Claude Mythos Preview as an invitation-only release under Project Glasswing, working with companies including Apple for defensive cybersecurity work: discovering and resolving software vulnerabilities before malicious actors exploit them. Access remains gated.

The Ghost in the Machine: Baba Said Inawezekana

While watching those Iranian Lego videos, I found myself thinking about a different one.I produced a Lego video of Baba, Raila Odinga, singing Lero ni Lero. It was the 2022 campaign anthem he recorded with Benga musician Emmanuel Musindi at FP Records in Kawangware. The message was direct: Leo ni Leo. Today is today. The one who says tomorrow is a liar. It became the anthem of the Azimio la Umoja movement

Baba did not make it to State House. The song outlived the campaign. The Chronicle, writing after his passing, described Lero ni Lero as part of Kenya’s cultural archive: “reminders of moments when melody met movement.”

Iran studied the American psyche for decades before making their videos. They used a Lego because it disarms. They used rap because it travels. Kenya already knows what Lero means. Inawezekana was not just a slogan. Lego Baba singing Lero is a memory. It is the kind of content Kenyans share because it touches something unfinished.

And here is the connection that matters now: the Kenyans who believed him are still here, still working, and now earning from the same global systems that once ignored this country.

Bright Brains in Kenya: The New Frontier

There is a bitter irony in being a Kenyan today. If you write in your own words, you often get flagged as “AI-generated.” It’s an irony worth naming: a significant portion of the training for these LLMs was done in Kenya. We trained the models. Now the models sound like us. Despite ongoing IP concerns, Kenyan talent is cracking the complex assessments used by global firms to identify top AI talent. On TikTok, a new subculture has emerged,mentors teaching thousands of others how to pass these high-stakes tests for platforms like Scale AI and Labelbox. One video said it simply: Jiamini tu. Believe in yourself. It is such an interesting time to be alive.

While the Pentagon signs $200$ million dollar contracts, there are Kenyans in bedsits in Roysambu and offices in Westlands landing jobs that pay $10,000$ USD per month in data annotation, model evaluation, and training work. They are the ones teaching AI to understand human nuance, math proofs, and legal logic. We are not just observers of this AI war; we are the human feedback that makes the technology possible.

Bridging the Gap: Hire the Source

The global demand for human-reviewed, culturally grounded data is growing. The talent is here. The education is here. The infrastructure is ready. If your organization requires high-quality annotation or RLHF (Reinforcement Learning from Human Feedback), you don’t need to look for a proxy in San Francisco or London.

Kenyans are already doing the work that powers the frontier models. If you want to engage this talent directly—ensuring both fair compensation and superior data quality—reach out:

Get in touch: suzie@susanngatia.africa

Jiamini Tu: The Toolkit for the New Era

If the world is going to be noisy, and if the tools of war and communication are going to be automated, the only response is to master the tools ourselves. The talent in Kenya is no longer waiting for “official operations.” We are building our own digital resilience.

If you are a creator or a dev in this space, here is your survival kit for the current landscape:

    • The Digital Presence: To overcome the IP disadvantage, use the Remote Work Setup Kit (approx. KES 2,500). Check out the details here…..
    • The Training Ground: Master Label Studio (open source) to build enterprise annotation skills. On the synthetic data side, tools like Gretel.ai or Mostly AI cover privacy-safe datasets relevant to healthcare and finance contracts.
    • The Content Engine: Use InVideo AI (text-to-video), ElevenLabs (voice, including Swahili), OpenArt (custom visuals), and Faceless.video (automated channel publishing) to tell our stories.

Full directory of these tools: susanngatia.africa/ai-tools-directory

Dev Roundup: What Moved This Week

Yann LeCun launched AMI Labs, Advanced Machine Intelligence, in March 2026. The Paris-based startup raised $1.03 billion at a $3.5 billion pre-money valuation, the largest seed round in European history, backed by Bezos Expeditions, NVIDIA, Temasek, and Samsung.⁸ AMI is building world models: AI systems designed to understand physical reality rather than predict text. CEO is Alexandre LeBrun. LeCun serves as executive chairman.

Seedance 2.0, from ByteDance, the parent company of TikTok, launched February 10, 2026. It generates up to 2K video with fully synchronized native audio in a single pass. It topped the Artificial Analysis video leaderboard with an Elo score of 1,269, beating Google Veo 3 and OpenAI Sora 2.⁹ Disney sent a cease and desist letter on February 13, 2026. US Senators Marsha Blackburn and Peter Welch wrote to ByteDance’s CEO demanding the model be shut down over copyright concerns.

Anthropic launched Claude Managed Agents in public beta on April 8, 2026, a cloud-hosted service for building and running autonomous AI agents. Billing is $0.08 per session hour plus standard API token prices.¹⁰ Notion, Rakuten, Asana, and Sentry are already in production. Anthropic says it cuts agent deployment from months to weeks.

Google released Gemma 4 on April 2, 2026, under an Apache 2.0 license. Four models: E2B, E4B, 26B Mixture of Experts, and 31B Dense. The 31B ranks third among all open models on Arena AI, outperforming Meta’s Llama 4 on math, coding, and reasoning at roughly one-twentieth the parameter count, with a 256K context window and support for over 140 languages.¹¹

A lot moved this week. The world is noisy, the tools are multiplying, and somewhere in Kenya, someone is cracking a $10,000 assessment.

Jiamini tu.


Notes

Get the Lastest in Communication- Media and the Latest tools

Categorized in: