In partnership with
In Today’s Issue:
🔐 EVMbench reveals that GPT-5.3-Codex can successfully exploit smart contract vulnerabilities
🎵 Lyria 3 is now live in the Gemini app for users 18+ globally
🎓 AI is creating a "Rust Belt moment" for office workers
👤 A new study finds that humans perform barely better than chance at spotting AI faces
✨ And more AI goodness…
Dear Readers,
Google just handed everyone with a Gemini account the power to produce studio-quality music from a text prompt - and if that doesn't make you rethink what "creative tools" means in 2025, today's issue will. We're breaking down DeepMind's Lyria 3 launch and why it could spell serious trouble for AI music startups like Suno.
But music is just the opening act: OpenAI's new EVMbench is stress-testing AI agents against real smart contract exploits (GPT-5.3-Codex already scores 72% on attack tasks), The Atlantic sounds the alarm on a white-collar unemployment crisis that government stimulus can't fix, and researchers found that even expert "super-recognisers" can barely tell AI-generated faces from real ones.
We've also got a fascinating deep dive into Endurance Bio's attempt to treat aging itself as a cellular bug worth patching, plus a striking graph showing Gemini 3 Deep Think demolishing the competition on LiveCodeBench Pro. Grab your coffee - this one's packed.
All the best,
🚀 EVMbench Elevates Smart Contract Security
Smart contracts safeguard over $100B+ in crypto assets, and EVMbench is here to stress-test AI agents in this high-stakes arena. Built with Paradigm, the benchmark evaluates AI performance across 120 real-world vulnerabilities in three modes—detect, patch, and exploit - revealing that GPT-5.3-Codex scores 72.2% in exploit tasks, a sharp jump from GPT-5’s 31.9% just six months ago.
While AI excels at draining funds in controlled exploit scenarios, it still struggles with exhaustive audits and fully secure patches—highlighting both progress and risk in AI-driven cybersecurity. Backed by $10M in API credits and ecosystem safeguards, EVMbench isn’t just a benchmark - it’s a call to strengthen AI-powered defenses before attackers catch up.
📉 AI Threatens White-Collar Stability
Annie Lowrey warns that AI-driven automation could trigger a structural crisis for white-collar workers, not just a typical recession. College graduates now make up a record 25% of the unemployed, and jobs most exposed to automation are seeing sharp spikes in layoffs - even as companies like Salesforce and KPMG cut staff and costs while deploying AI. Unlike past downturns, government stimulus may not fix a labor market where businesses simply don’t need displaced accountants, engineers, or managers anymore.
If AI eliminates large swaths of office work, the U.S. could face long-term structural unemployment, rising inequality, and weakened social mobility, echoing the industrial decline of Rust Belt cities decades ago. Proposals like a $1,500/month universal basic income, championed by leaders such as Sam Altman, may soften the blow, but risk creating a society detached from the meaning and stability work provides. The big question: can the economy reinvent white-collar work fast enough to avoid a 30% unemployment nightmare?
🧠 AI Faces Fool Most People
A new study from UNSW Sydney and Australian National University, published in the British Journal of Psychology, reveals that 125 participants - including 36 “super-recognisers” - barely performed better than chance when spotting AI-generated faces. Even experts were fooled, despite being highly confident in their judgments.
Today’s most advanced AI faces aren’t flawed - they’re “too perfect,” with high symmetry and average proportions that make them statistically typical and incredibly convincing. The takeaway: visual judgment alone is no longer reliable, raising real risks for scams, fake profiles, and identity fraud across social media, dating, and hiring.
“Only a Small Number of Years” - Anthropic CEO Says AI Will Surpass Humans Soon
Google AI Makes Music Now
The Takeaway
👉 Google launched Lyria 3 inside the Gemini app, making AI music generation available to all users 18+ globally - a major shift from limited creator-only access
👉 The model generates complete 30-second tracks with vocals, lyrics, and cover art from simple text prompts or uploaded images, with control over style, tempo, and vocals
👉 All generated tracks carry invisible SynthID watermarks, and Gemini can now verify whether any uploaded audio was AI-generated - a critical step for content authentication
👉 The move puts direct competitive pressure on AI music startups like Suno and signals that Big Tech sees music generation as the next major consumer AI feature
The AI race just got a soundtrack. Google DeepMind dropped Lyria 3 - and it might just change how we think about music creation forever.
Here's the deal: Lyria 3 is Google's most advanced music generation model, now available inside the Gemini app. Type a prompt; say, a funky jazz piece with Brazilian vocals - or upload a photo, and the AI creates a 30-second high-fidelity track complete with lyrics, vocals, and cover art. Unlike simpler models that piece together loops, Lyria 3 generates full musical arrangements from scratch, handling melody, harmony, rhythm, and timbre simultaneously. Users can fine-tune style, tempo, and vocal characteristics with surprising precision.
This is the first version of Lyria to get a wide public release, previous models were limited to select musicians and YouTube creators. Now it's available globally in eight languages. All outputs are watermarked with SynthID technology for AI content identification. And with competitors like Suno having raised $250 million, this launch signals Google is going all-in on the AI music generation space.
Will this increase the pressure on Suno – or even be the breakthrough for AI music?
Why it matters: Google is democratizing music production at scale, making professional-grade audio creation accessible to anyone with a Gemini account. This puts immense pressure on AI music startups and could fundamentally reshape how creators produce soundtracks, content, and art.
Sources:
🔗 https://deepmind.google/models/lyria/
Write like a founder, faster
When the calendar is full, fast, clear comms matter. Wispr Flow lets founders dictate high-quality investor notes, hiring messages, and daily rundowns and get paste-ready writing instantly. It keeps your voice and the nuance you rely on for strategic messages while removing filler and cleaning punctuation. Save repeated snippets to scale consistent leadership communications. Works across Mac, Windows, and iPhone. Try Wispr Flow for founders.
AI Meets Anti-Aging Medicine
A tiny protein could hold the secret to how we age, and a small biotech just moved one step closer to proving it. Endurance Bio, a clinical-stage company born out of Stanford research, is preparing Phase 2 trials for a drug that targets PGC-1α, a protein scientists call the "master regulator" of cell energy metabolism. Think of it this way: when our cells age, their power plants (mitochondria) break down and their waste disposal systems (lysosomes) clog up. Endurance's compound T-168 works by boosting PGC-1α and TFEB to restore both mitochondrial and lysosomal function - essentially rebooting the cell's entire energy and cleanup infrastructure.
The company's first target is Parkinson's disease, where over 70 randomized clinical trials have failed to deliver a compound capable of halting disease progression. But co-founder Frederic Godderis sees much bigger potential: animal studies have shown promising results in ALS, type 2 diabetes, muscle degeneration, and general age-related frailty. A second molecule, T-621, is already in preclinical studies, and the team has a full library of compounds waiting in the pipeline.
Endurance Bio is one of the first companies to bring a PGC-1α-targeting drug to Phase 2 human trials, potentially opening an entirely new therapeutic approach for neurodegenerative and aging-related diseases. If the clinical data hold up, this could redefine how medicine approaches aging itself , not as inevitable decline, but as a treatable cellular dysfunction.
The best marketing ideas come from marketers who live it. That’s what The Marketing Millennials delivers: real insights, fresh takes, and no fluff. Written by Daniel Murray, a marketer who knows what works, this newsletter cuts through the noise so you can stop guessing and start winning. Subscribe and level up your marketing game.
