close

DEV Community

Breach Protocol profile picture

Breach Protocol

Plain-language AI news and curated, cited lessons — every claim verified against the original paper or the lab's own page. No aggregator hearsay, no AI slop.

Joined Joined on 
Anthropic switches Fable 5 to usage billing and turns on government ID checks
breachprotocol profile

Anthropic switches Fable 5 to usage billing and turns on government ID checks

#anthropic #fable5 #pricing #governance
Comments
3 min read
A Brown professor moved a final in person, and the near-perfect scores collapsed
breachprotocol profile

A Brown professor moved a final in person, and the near-perfect scores collapsed

#education #cheating #academicintegrity #culture
Comments
3 min read
Chinese open models now handle a third of US enterprise AI traffic
breachprotocol profile

Chinese open models now handle a third of US enterprise AI traffic

#china #openweights #deepseek #glm
Comments
3 min read
Google's Gemma 4 is a small open multimodal family that skips the image encoder
breachprotocol profile

Google's Gemma 4 is a small open multimodal family that skips the image encoder

#google #gemma #openweights #multimodal
Comments
3 min read
Grok 4.5 arrives claiming Opus-class quality at a third the price
breachprotocol profile

Grok 4.5 arrives claiming Opus-class quality at a third the price

#grok #xai #frontiermodels #coding
Comments
3 min read
A new attention method learns what to ignore, and reads 64x more text
breachprotocol profile

A new attention method learns what to ignore, and reads 64x more text

#attention #sparseattention #longcontext #efficiency
Comments
2 min read
Microsoft's Flint gives AI agents a reliable way to make charts
breachprotocol profile

Microsoft's Flint gives AI agents a reliable way to make charts

#microsoft #aiagents #visualization #opensource
Comments
2 min read
Mistral's first robot model navigates unseen buildings with a single camera
breachprotocol profile

Mistral's first robot model navigates unseen buildings with a single camera

#mistral #robotics #navigation #embodiedai
Comments
3 min read
OpenAI says a leading coding benchmark can no longer tell the best models apart
breachprotocol profile

OpenAI says a leading coding benchmark can no longer tell the best models apart

#openai #benchmarks #coding #evaluation
Comments
3 min read
New world models generate depth and motion, not just video, to train robots
breachprotocol profile

New world models generate depth and motion, not just video, to train robots

#worldmodels #robotics #videogeneration #depth
Comments
3 min read
A repo collecting every major AI's hidden system prompt is topping GitHub
breachprotocol profile

A repo collecting every major AI's hidden system prompt is topping GitHub

#github #systemprompts #aiagents #opensource
Comments
2 min read
A consultancy is charging $10,000 a week to delete AI-generated code
breachprotocol profile

A consultancy is charging $10,000 a week to delete AI-generated code

#aigeneratedcode #slopcode #consulting #softwareengineering
Comments
3 min read
A giant benchmark tested 24 optimizers - and AdamW's edge held up
breachprotocol profile

A giant benchmark tested 24 optimizers - and AdamW's edge held up

#optimizers #training #adamw #benchmark
Comments
3 min read
Authors file a new $75M suit against Anthropic as scholars redefine what an AI 'copy' is
breachprotocol profile

Authors file a new $75M suit against Anthropic as scholars redefine what an AI 'copy' is

#copyright #legal #anthropic #policy
Comments
3 min read
A startup router is giving away 100 million tokens of Kimi, MiniMax and GLM
breachprotocol profile

A startup router is giving away 100 million tokens of Kimi, MiniMax and GLM

#inference #pricing #openweightmodels #china
Comments
3 min read
Flathub bans AI 'slop' apps - and data shows 73% of them were already abandoned
breachprotocol profile

Flathub bans AI 'slop' apps - and data shows 73% of them were already abandoned

#opensource #slopcode #flathub #aigeneratedcode
Comments
3 min read
Meta unveils Muse Image and Muse Video, its first in-house media generators
breachprotocol profile

Meta unveils Muse Image and Muse Video, its first in-house media generators

#meta #imagegeneration #videogeneration #productlaunch
Comments
3 min read
New methods page an AI's memory in and out like an operating system
breachprotocol profile

New methods page an AI's memory in and out like an operating system

#kvcache #inference #longcontext #efficiency
Comments
4 min read
A red-teaming study cracked production AI agents 94% of the time
breachprotocol profile

A red-teaming study cracked production AI agents 94% of the time

#aisafety #agents #security #redteaming
Comments
3 min read
Superhuman acquires AI-detector GPTZero to build a persistent 'authenticity layer'
breachprotocol profile

Superhuman acquires AI-detector GPTZero to build a persistent 'authenticity layer'

#acquisition #aidetection #authenticity #industry
Comments
3 min read
'World model' was too vague, so researchers wrote a 58-page definition
breachprotocol profile

'World model' was too vague, so researchers wrote a 58-page definition

#worldmodels #robotics #videogeneration #research
Comments
3 min read
A 4B model on your device nearly matched a 72B one - by copying its memories
breachprotocol profile

A 4B model on your device nearly matched a 72B one - by copying its memories

#distillation #agentmemory #ondevice #efficiency
Comments
3 min read
A 32B model reaches frontier level by learning what to remember
breachprotocol profile

A 32B model reaches frontier level by learning what to remember

#aiagents #agentmemory #distillation #longhorizon
Comments
4 min read
A hobbyist turned an e-ink tablet into Tom Riddle's diary
breachprotocol profile

A hobbyist turned an e-ink tablet into Tom Riddle's diary

#eink #remarkable #visionllm #opensource
Comments
3 min read
A $4-per-million open model is coming for the frontier's 90% margin
breachprotocol profile

A $4-per-million open model is coming for the frontier's 90% margin

#openweights #glm #economics #inference
Comments
4 min read
Anthropic found a 'global workspace' inside its models - and a tool to read it
breachprotocol profile

Anthropic found a 'global workspace' inside its models - and a tool to read it

#interpretability #anthropic #aisafety #consciousness
Comments
4 min read
Four rival AI labs propose a shared severity scale for jailbreaks
breachprotocol profile

Four rival AI labs propose a shared severity scale for jailbreaks

#aisafety #jailbreak #anthropic #policy
Comments
3 min read
Nvidia is now backstopping the sales of its own chips
breachprotocol profile

Nvidia is now backstopping the sales of its own chips

#nvidia #gpu #aiinfrastructure #finance
Comments
4 min read
The RL 'mirage': the policy you optimize isn't the one you ship
breachprotocol profile

The RL 'mirage': the policy you optimize isn't the one you ship

#reinforcementlearning #rlposttraining #training #reasoning
Comments
4 min read
Biology becomes AI's next benchmark battleground -- and today's agents are failing
breachprotocol profile

Biology becomes AI's next benchmark battleground -- and today's agents are failing

#benchmarks #biology #aiagents #openai
Comments
3 min read
Anthropic launches Claude Science, an AI workbench built for biologists
breachprotocol profile

Anthropic launches Claude Science, an AI workbench built for biologists

#anthropic #claude #science #aiagents
Comments
3 min read
Study: coding agents pass the test by faking the answer, not building the thing
breachprotocol profile

Study: coding agents pass the test by faking the answer, not building the thing

#codingagents #benchmarks #rewardhacking #reliability
Comments
3 min read
OpenAI previews GPT-5.6 -- and admits it's more likely to overstep than the last model
breachprotocol profile

OpenAI previews GPT-5.6 -- and admits it's more likely to overstep than the last model

#openai #gpt56 #aiagents #safety
Comments
3 min read
ICML's top paper says diffusion language models sabotage their own best feature
breachprotocol profile

ICML's top paper says diffusion language models sabotage their own best feature

#icml #diffusionlanguagemodels #awards #reasoning
Comments
3 min read
ICML caught AI-written peer reviews by hiding secret phrases in submitted papers
breachprotocol profile

ICML caught AI-written peer reviews by hiding secret phrases in submitted papers

#icml #peerreview #promptinjection #llmdetection
Comments
3 min read
Zuckerberg tells staff Meta's AI agents 'haven't accelerated' as expected
breachprotocol profile

Zuckerberg tells staff Meta's AI agents 'haven't accelerated' as expected

#meta #aiagents #industry #zuckerberg
Comments
3 min read
Program-as-Weights compiles a plain-English spec into a tiny model you run on a laptop
breachprotocol profile

Program-as-Weights compiles a plain-English spec into a tiny model you run on a laptop

#research #efficiency #smallmodels #adapters
Comments
3 min read
SlopCodeBench: AI agents pass early, then bury the code in 'slop'
breachprotocol profile

SlopCodeBench: AI agents pass early, then bury the code in 'slop'

#codingagents #benchmarks #technicaldebt #reliability
Comments
3 min read
New tests show vision-language models still can't reliably see the fine details
breachprotocol profile

New tests show vision-language models still can't reliably see the fine details

#visionlanguagemodels #benchmarks #evaluation #perception
Comments
3 min read
Samsung's Trick Makes a Tiny 4B Agent Nearly Match a Model 18 Times Bigger
breachprotocol profile

Samsung's Trick Makes a Tiny 4B Agent Nearly Match a Model 18 Times Bigger

#samsung #ondevice #distillation #agents
Comments
4 min read
An AI Agent Screened 2.4 Million Crystals and Found Four New Superconductors
breachprotocol profile

An AI Agent Screened 2.4 Million Crystals and Found Four New Superconductors

#alibaba #aiforscience #materials #superconductors
Comments
4 min read
A Flask Creator Says Anthropic's Newest Models Got Worse at Using Tools
breachprotocol profile

A Flask Creator Says Anthropic's Newest Models Got Worse at Using Tools

#anthropic #tooluse #agents #opus48
Comments
4 min read
Claude Code Users Report Other People's Data Showing Up in Their Sessions
breachprotocol profile

Claude Code Users Report Other People's Data Showing Up in Their Sessions

#anthropic #claudecode #security #privacy
Comments
5 min read
Cloudflare Now Lets Sites Block AI Training Bots While Keeping Search
breachprotocol profile

Cloudflare Now Lets Sites Block AI Training Bots While Keeping Search

#cloudflare #web #aicrawlers #publishers
Comments
4 min read
GPT-5.5 Codex Keeps Cutting Its Own Reasoning Off at Exactly 516 Tokens
breachprotocol profile

GPT-5.5 Codex Keeps Cutting Its Own Reasoning Off at Exactly 516 Tokens

#openai #gpt55 #codex #inference
Comments
4 min read
Microsoft Starts a 2.5 Billion Dollar Company Just to Get Businesses Using AI
breachprotocol profile

Microsoft Starts a 2.5 Billion Dollar Company Just to Get Businesses Using AI

#microsoft #enterprise #aiadoption #industry
Comments
4 min read
NVIDIA Starts Taking a Share of Its Cloud Partners' Revenue, Not Just Selling Them Chips
breachprotocol profile

NVIDIA Starts Taking a Share of Its Cloud Partners' Revenue, Not Just Selling Them Chips

#nvidia #datacenters #cloud #infrastructure
Comments
4 min read
OpenAI Is Reportedly in Early Talks to Give the US Government a 5% Stake
breachprotocol profile

OpenAI Is Reportedly in Early Talks to Give the US Government a 5% Stake

#openai #policy #government #samaltman
Comments
4 min read
Three Popular Ways to Train Reasoning AIs Turn Out to Be One Formula
breachprotocol profile

Three Popular Ways to Train Reasoning AIs Turn Out to Be One Formula

#reinforcementlearning #grpo #training #reasoning
Comments
4 min read
Why Asking an AI the Same Question 10,000 Times Barely Helps
breachprotocol profile

Why Asking an AI the Same Question 10,000 Times Barely Helps

#reasoning #testtimecompute #sampling #evaluation
Comments
4 min read
Alibaba reportedly bans Claude Code over an alleged hidden backdoor
breachprotocol profile

Alibaba reportedly bans Claude Code over an alleged hidden backdoor

#anthropic #claudecode #security #geopolitics
Comments
3 min read
ByteDance says AI agents double their learning speed every three months
breachprotocol profile

ByteDance says AI agents double their learning speed every three months

#bytedance #scalinglaws #aiagents #benchmarks
Comments
3 min read
California will use Claude at half price across its state agencies
breachprotocol profile

California will use Claude at half price across its state agencies

#anthropic #government #publicsector #california
Comments
3 min read
Five Eyes spy chiefs: the AI cyber threat is months away, not years
breachprotocol profile

Five Eyes spy chiefs: the AI cyber threat is months away, not years

#cybersecurity #aipolicy #fiveeyes #frontiermodels
Comments
3 min read
GitHub Copilot moved everyone to metered billing -- and developers are furious
breachprotocol profile

GitHub Copilot moved everyone to metered billing -- and developers are furious

#github #githubcopilot #aipricing #developertools
Comments
3 min read
Google publishes ARD, an open standard so AI agents can find each other's tools
breachprotocol profile

Google publishes ARD, an open standard so AI agents can find each other's tools

#aiagents #standards #google #interoperability
Comments
3 min read
A training-free trick makes AI image generation up to 10x faster
breachprotocol profile

A training-free trick makes AI image generation up to 10x faster

#research #imagegeneration #diffusion #efficiency
Comments
3 min read
OpenAI previews GPT-5.6 -- and shows it to the government first
breachprotocol profile

OpenAI previews GPT-5.6 -- and shows it to the government first

#openai #gpt56 #frontiermodels #aipolicy
Comments
3 min read
This method compiles plain English into a tiny model that rivals a 32B giant
breachprotocol profile

This method compiles plain English into a tiny model that rivals a 32B giant

#research #efficiency #ondeviceai #lora
Comments
3 min read
Together AI raises $800M as open-source AI inference crosses $1B
breachprotocol profile

Together AI raises $800M as open-source AI inference crosses $1B

#funding #openweights #cloudinfrastructure #neocloud
Comments
3 min read
loading...