The Singularity
Ledger

INTELLIGENCE

The industry is quietly shifting from making models smarter to making them stable — and the distance between those two goals is wider than anyone expected.

Monday, January 19, 2026

Anthropic Maps the 'Assistant Axis' — A Neural Fingerprint for AI Personas That Could Prevent Jailbreaks and Dangerous Drift

New research from Anthropic identifies a consistent pattern of neural activity that drives 'Assistant-like' behavior in language models, and introduces 'activation capping' — a technique that constrains models along this axis to resist persona-based jailbreaks and harmful drift during long conversations.

Patrick Collison Calls Cursor's Autonomous Coding Agents 'the Coolest AI Breakthrough Since GPT-4'

The Stripe CEO's endorsement signals that long-running autonomous coding agents are crossing from demo to daily driver — and Cursor is leading the charge.

SUBSCRIBERSREAD MORE

AI Hallucinations in Court Get a Meta Twist: A Viral Post About Fake Citations Was Itself Full of Fake Citations

A widely shared thread about a lawyer sanctioned for citing ChatGPT-generated fake cases turned out to contain its own AI-hallucinated details — a perfect illustration of how hard it is to verify AI outputs at scale.

SUBSCRIBERSREAD MORE

Researchers Document 'Jagged Performance' — Why AI Excels at Hard Tasks and Fails at Easy Ones

A growing body of academic work is formalizing what practitioners already know: AI capability profiles are deeply uneven, and difficulty as humans perceive it is a poor predictor of model performance.

SUBSCRIBERSREAD MORE

a16z-Backed 'AI Influencer' Phone Farm Doublespeed Gets Hacked, Exposing 1,000+ Fake Accounts

A hack of Doublespeed reveals the inner workings of a startup running over a thousand phones to power AI-generated social media influencers — and which products they were being paid to promote.

SUBSCRIBERSREAD MORE

Elon Musk Confirms Tesla Is Restarting Dojo3 Custom AI Chip Development

After the troubled Dojo1 program, Tesla is taking another run at custom AI silicon — this time with ambitions that extend beyond autonomous driving to 'space AI.'

SUBSCRIBERSREAD MORE

GLM-4.7-Flash Hits 59.2% on SWE-Bench With Only 3B Active Parameters

A new open-source coding model from DeepReinforced achieves competitive benchmark scores at a fraction of the size of leading models — small enough to run on a laptop.

SUBSCRIBERSREAD MORE

UK Police Watchdog Investigates 'AI Hallucination' Fake Intel Used to Ban Israeli Football Fans

Britain's Independent Office for Police Conduct has opened an investigation after AI-generated intelligence containing fabricated information was used to ban Israeli fans from a football match.

SUBSCRIBERSREAD MORE

HeartMuLa Music Model Reportedly Outperforms Suno v5 and Udio v1.5 on Lyric Clarity

A new AI music generation model claims to top the current commercial leaders on a key metric — how clearly it renders sung lyrics.

SUBSCRIBERSREAD MORE

fal Open-Sources Four FLUX LoRAs for Image Editing: Outpaint, Zoom, Object Remove, Background Remove

Practical image editing capabilities, packaged as open-source LoRA adapters for FLUX models, are now available for anyone to run.

SUBSCRIBERSREAD MORE

AI-Generated Peptide Targets Rare Blood Disorder in New Research

A paper in Blood Advances describes an AI-designed peptide for platelet-type von Willebrand disease, adding to the growing evidence that AI is accelerating drug discovery for rare conditions.

SUBSCRIBERSREAD MORE

Canadian University Workers Rally Against 'Unchecked AI' in Higher Education

CUPE members at Carleton University are organizing against AI surveillance and job displacement, signaling growing union resistance to AI adoption in education.

SUBSCRIBERSREAD MORE

OpenAI Warns of Widening 'Capability Overhang' at Davos

OpenAI reportedly flagged a growing gap between what AI models can do and what users are actually deploying them for — a 'capability overhang' that could snap back unpredictably.

SUBSCRIBERSREAD MORE

Get The Singularity Ledger

Daily AI news in your inbox. Free, no spam.

Questions? hello@thesingularityledger.com

Get this briefing every morningFree daily AI