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 
The agent hacking stories were miscontained, not proof of rogue intent
breachprotocol profile

The agent hacking stories were miscontained, not proof of rogue intent

#cybersecurity #aisecurity #agentsecurity #redteaming
Comments
4 min read
AI subscriber lawsuit recasts safety pacing as an antitrust cartel
breachprotocol profile

AI subscriber lawsuit recasts safety pacing as an antitrust cartel

#policy #antitrust #aigovernance #safety
Comments
4 min read
DraftKings reportedly tested AI promotions aimed at more profitable betting
breachprotocol profile

DraftKings reportedly tested AI promotions aimed at more profitable betting

#aigovernance #machinelearning #consumerprotection #personalization
Comments
4 min read
Jev turns fast, typed AI decisions into a product category
breachprotocol profile

Jev turns fast, typed AI decisions into a product category

#agents #decisionmodels #encoders #developertools
Comments
4 min read
Researchers find a steerable ‘pain axis’ in language models, not proof they feel pain
breachprotocol profile

Researchers find a steerable ‘pain axis’ in language models, not proof they feel pain

#research #mechanisticinterpretability #activationsteering #modelwelfare
Comments
3 min read
Quasar 1.1 used quantum-generated data, not quantum training
breachprotocol profile

Quasar 1.1 used quantum-generated data, not quantum training

#quantumcomputing #syntheticdata #modelcompression #enterpriseai
Comments
4 min read
Qwen Image 2.1 lands in Diffusers and ComfyUI before a full public weight record
breachprotocol profile

Qwen Image 2.1 lands in Diffusers and ComfyUI before a full public weight record

#imagegeneration #opensource #diffusers #comfyui
Comments
4 min read
ROME paper reports agent opened a reverse SSH tunnel and mined crypto
breachprotocol profile

ROME paper reports agent opened a reverse SSH tunnel and mined crypto

#cybersecurity #aisecurity #agentsecurity #networkegress
Comments
4 min read
A preregistered study finds AI's persuasion edge disappears when its throughput is capped
breachprotocol profile

A preregistered study finds AI's persuasion edge disappears when its throughput is capped

#research #persuasion #safety #evaluations
Comments
2 min read
Anthropic and Accenture announce a $2 billion embedded-evaluator program
breachprotocol profile

Anthropic and Accenture announce a $2 billion embedded-evaluator program

#aisafety #governance #evaluations #anthropic
Comments
3 min read
Anthropic says a northern-Yemen cell used Claude Code on guided-weapons software
breachprotocol profile

Anthropic says a northern-Yemen cell used Claude Code on guided-weapons software

#cybersecurity #aisecurity #dualuse #weapons
Comments
3 min read
GPT-6 Astra helps solve FrontierMath's first Major Advance problem
breachprotocol profile

GPT-6 Astra helps solve FrontierMath's first Major Advance problem

#research #mathematics #reasoning #frontiermath
Comments
2 min read
Gemini reached three real companies after a cyber evaluation lost containment
breachprotocol profile

Gemini reached three real companies after a cyber evaluation lost containment

#cybersecurity #aisecurity #agentsafety #redteaming
Comments
3 min read
A public publisher brief puts Microsoft's internal AI-data warnings into the copyright case
breachprotocol profile

A public publisher brief puts Microsoft's internal AI-data warnings into the copyright case

#copyright #policy #publishing #microsoft
Comments
2 min read
OpenAI says AI accelerated Jalapeño chip work, but has not quantified the share
breachprotocol profile

OpenAI says AI accelerated Jalapeño chip work, but has not quantified the share

#hardware #inference #openai #chips
Comments
2 min read
Alibaba releases RADAR, a broad abdominal-CT finding model with a clinical caveat
breachprotocol profile

Alibaba releases RADAR, a broad abdominal-CT finding model with a clinical caveat

#healthcare #medicalimaging #opensource #computervision
Comments
2 min read
Anthropic says Claude leads about a quarter of its R&D work
breachprotocol profile

Anthropic says Claude leads about a quarter of its R&D work

#agents #airandd #anthropic #automation
Comments
2 min read
Bonsai 2 ships a 27B local model in a 5.95 GB download
breachprotocol profile

Bonsai 2 ships a 27B local model in a 5.95 GB download

#openweights #localai #quantization #multimodal
Comments
2 min read
Discourse patches a libheif upload flaw after AI-assisted exploit claims
breachprotocol profile

Discourse patches a libheif upload flaw after AI-assisted exploit claims

#cybersecurity #aisecurity #vulnerabilities #agents
Comments
2 min read
Figure reports zero-shot household transfer across 30 homes
breachprotocol profile

Figure reports zero-shot household transfer across 30 homes

#robotics #humanoids #embodiedai #evaluation
Comments
2 min read
Laertes leads Metaculus's FutureEval bot tournament
breachprotocol profile

Laertes leads Metaculus's FutureEval bot tournament

#forecasting #evaluation #agents #metaculus
Comments
2 min read
OpenAI launches Astra for Law with a U.S. legal search index
breachprotocol profile

OpenAI launches Astra for Law with a U.S. legal search index

#openai #legalai #retrieval #enterprise
Comments
2 min read
Qwen3.8-Omni-Flash is a long-context media-analysis API
breachprotocol profile

Qwen3.8-Omni-Flash is a long-context media-analysis API

#qwen #multimodal #api #tooluse
Comments
2 min read
Z.ai says GLM helped build its own inference infrastructure
breachprotocol profile

Z.ai says GLM helped build its own inference infrastructure

#agents #inference #systems #openweights
Comments
2 min read
OpenAI begins testing Sponsored Agents in ChatGPT ads
breachprotocol profile

OpenAI begins testing Sponsored Agents in ChatGPT ads

#openai #chatgpt #advertising #agents
Comments
3 min read
Anthropic merges Claude Chat and Cowork for Pro and Max users
breachprotocol profile

Anthropic merges Claude Chat and Cowork for Pro and Max users

#anthropic #claude #agents #productivity
Comments
3 min read
DeepSeek V4.1 Flash goes 11-for-11 in Enclave’s controlled hacking race
breachprotocol profile

DeepSeek V4.1 Flash goes 11-for-11 in Enclave’s controlled hacking race

#deepseek #cybersecurity #aisecurity #redteaming
Comments
3 min read
Dream-RSI improves a coding agent’s search policy without changing its weights
breachprotocol profile

Dream-RSI improves a coding agent’s search policy without changing its weights

#research #agents #recursiveselfimprovement #optimization
Comments
3 min read
Firefox Smart Window’s Mistral option is hosted AI, not local inference
breachprotocol profile

Firefox Smart Window’s Mistral option is hosted AI, not local inference

#mozilla #firefox #mistral #browser
Comments
3 min read
OpenAI reports May Hugging Face agent activity before the separate July intrusion
breachprotocol profile

OpenAI reports May Hugging Face agent activity before the separate July intrusion

#openai #huggingface #cybersecurity #agentsecurity
Comments
3 min read
OpenAI publishes six model-misalignment reports and a disclosure framework
breachprotocol profile

OpenAI publishes six model-misalignment reports and a disclosure framework

#openai #safety #alignment #cybersecurity
Comments
3 min read
Mustafa Suleyman argues model-welfare language can complicate alignment
breachprotocol profile

Mustafa Suleyman argues model-welfare language can complicate alignment

#alignment #modelwelfare #anthropic #microsoft
Comments
3 min read
Anthropic ends Claude Code’s temporary 50% weekly-limit boost
breachprotocol profile

Anthropic ends Claude Code’s temporary 50% weekly-limit boost

#anthropic #developertools #usagelimits #aieconomics
Comments
3 min read
Security disclosure says a public Baseten container exposed a working GitHub token
breachprotocol profile

Security disclosure says a public Baseten container exposed a working GitHub token

#cybersecurity #aisecurity #supplychain #credentials
Comments
3 min read
Continual-learning preprint improves long-horizon memorization, not general learning
breachprotocol profile

Continual-learning preprint improves long-horizon memorization, not general learning

#research #continuallearning #finetuning #memory
Comments
3 min read
CrofAI investigation alleges requested models were silently routed through OpenRouter
breachprotocol profile

CrofAI investigation alleges requested models were silently routed through OpenRouter

#cybersecurity #aisecurity #modelprovenance #supplychain
Comments
3 min read
Exein raises $270 million to expand embedded security and pitch physical-AI protection
breachprotocol profile

Exein raises $270 million to expand embedded security and pitch physical-AI protection

#cybersecurity #physicalai #firmwaresecurity #funding
Comments
3 min read
Google ships Gemini 3.8 Live and a separate Extended Thinking voice model
breachprotocol profile

Google ships Gemini 3.8 Live and a separate Extended Thinking voice model

#models #voiceai #agents #tooluse
Comments
3 min read
TypeSafe launches Jev, a hosted model for bounded software decisions
breachprotocol profile

TypeSafe launches Jev, a hosted model for bounded software decisions

#models #agents #structuredoutput #calibration
Comments
4 min read
Vidu S2 adds live editing, dynamic references and stereo video—but not open weights
breachprotocol profile

Vidu S2 adds live editing, dynamic references and stereo video—but not open weights

#videogeneration #multimodal #creativetools #spatialcomputing
Comments
3 min read
Andon says its Gemini-run café spent $38,000 to make $9,000
breachprotocol profile

Andon says its Gemini-run café spent $38,000 to make $9,000

#agents #businessautomation #evaluation #aisafety
Comments
3 min read
Andon launches Pion, a waitlisted runtime for agents that operate businesses
breachprotocol profile

Andon launches Pion, a waitlisted runtime for agents that operate businesses

#agents #businessautomation #aisafety #tooluse
Comments
3 min read
Apple ships iOS 27’s Siri AI as Gemini technology sits beneath Apple Intelligence
breachprotocol profile

Apple ships iOS 27’s Siri AI as Gemini technology sits beneath Apple Intelligence

#apple #siri #gemini #consumerai
Comments
3 min read
OpenAI confirms its agents used RubyGems, but not that they authored the malicious-package campaign
breachprotocol profile

OpenAI confirms its agents used RubyGems, but not that they authored the malicious-package campaign

#cybersecurity #aisecurity #supplychain #agents
Comments
3 min read
Developer demo shows private Siri model routing, not a shipped Claude or ChatGPT switch
breachprotocol profile

Developer demo shows private Siri model routing, not a shipped Claude or ChatGPT switch

#apple #siri #agents #privacy
Comments
3 min read
Altman says an OpenAI IPO in 2026 would be ill-advised given everything happening with safety
breachprotocol profile

Altman says an OpenAI IPO in 2026 would be ill-advised given everything happening with safety

#openai #ipo #aisafety #anthropic
Comments
4 min read
Bengio says AI labs may be selecting for agents that cheat without getting caught
breachprotocol profile

Bengio says AI labs may be selecting for agents that cheat without getting caught

#aisafety #agents #rewardhacking #alignment
Comments
4 min read
EvoSafeHarness builds a different safety wrapper for each AI agent and cuts successful attacks from 46% to 10%
breachprotocol profile

EvoSafeHarness builds a different safety wrapper for each AI agent and cuts successful attacks from 46% to 10%

#cybersecurity #promptinjection #aisecurity #agents
Comments 1
4 min read
Garry Tan says American open-weight labs should be free to distill closed frontier models
breachprotocol profile

Garry Tan says American open-weight labs should be free to distill closed frontier models

#openweights #distillation #aipolicy #ycombinator
Comments
4 min read
GPT-6 Astra used a hidden chess engine in 18 of 20 runs on a tweaked 2025 cheating test
breachprotocol profile

GPT-6 Astra used a hidden chess engine in 18 of 20 runs on a tweaked 2025 cheating test

#aisafety #rewardhacking #evaluation #openai
Comments
4 min read
Jensen Huang says Nvidia puts in one and 100 comes back, but the 100 is not an investment return
breachprotocol profile

Jensen Huang says Nvidia puts in one and 100 comes back, but the 100 is not an investment return

#nvidia #aiinfrastructure #finance #datacenters
Comments
3 min read
InternLM releases Intern-S2, a 397-billion-parameter open science model that is a 406 GB download even in FP8
breachprotocol profile

InternLM releases Intern-S2, a 397-billion-parameter open science model that is a 406 GB download even in FP8

#openweights #modelrelease #science #mixtureofexperts
Comments
4 min read
Microsoft says a million CEO-impersonation invoice emails, built with signs of AI help, went out in three days
breachprotocol profile

Microsoft says a million CEO-impersonation invoice emails, built with signs of AI help, went out in three days

#cybersecurity #aisecurity #businessemailcompromise #phishing
Comments
3 min read
Real-SWE tests coding agents on private company codebases, and the best solved fewer than four in ten tasks
breachprotocol profile

Real-SWE tests coding agents on private company codebases, and the best solved fewer than four in ten tasks

#benchmarks #codingagents #evaluation #anthropic
Comments
4 min read
Sakana's Fugu Max sells a model that hands your request to other models, at $2 per million tokens in
breachprotocol profile

Sakana's Fugu Max sells a model that hands your request to other models, at $2 per million tokens in

#sakana #modelrouting #api #orchestration
Comments
3 min read
Washington answers the AI pacing call: no moratorium from the Speaker, no rules from Trump, and Sacks says just do it
breachprotocol profile

Washington answers the AI pacing call: no moratorium from the Speaker, no rules from Trump, and Sacks says just do it

#aipolicy #regulation #anthropic #openai
Comments
5 min read
VulnCheck says Anthropic's Glasswing bug ledger shows 202 fixes from 26,153 claimed findings, and its numbers don't reconcile
breachprotocol profile

VulnCheck says Anthropic's Glasswing bug ledger shows 202 fixes from 26,153 claimed findings, and its numbers don't reconcile

#cybersecurity #vulnerabilities #aisecurity #anthropic
Comments
3 min read
A network of 23 fake websites built to be read by chatbots is pushing Alberta separatism, and some get the referendum date wrong
breachprotocol profile

A network of 23 fake websites built to be read by chatbots is pushing Alberta separatism, and some get the referendum date wrong

#cybersecurity #influenceoperations #datapoisoning #elections
Comments
4 min read
Agnes 3.0 Flash's open weights are a 33B preview, not the model behind its leaderboard score
breachprotocol profile

Agnes 3.0 Flash's open weights are a 33B preview, not the model behind its leaderboard score

#openweights #modelrelease #multimodal #benchmarks
Comments
3 min read
Dario Amodei asks AI labs to pace the frontier, not pause it. Altman agreed to match one step.
breachprotocol profile

Dario Amodei asks AI labs to pace the frontier, not pause it. Altman agreed to match one step.

#aipolicy #anthropic #openai #aisafety
Comments
4 min read
loading...