Habr AI
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
YADRO bans its LLM from touching numbers and ships 10+ reports a day with no humans in the loop
In three months, YADRO analysts built a service on Dify, Trino and Airflow. Code computes the metrics, the model writes commentary with placeholders, and over a dozen reports go out each morning.
Security engineer builds an MCP server in 7 days that spooks an AI agent with vulnerabilities
AppSecGate MCP runs scanners and keeps a findings database. On OWASP Juice Shop, an agent worked through 200 findings in 50 minutes for 15 rubles and fixed 56, including 8 critical.
aiedit hides individual lines of code from an AI agent: 5 levels and 857 tests
A Python beginner built the aiedit editor with Claude: the model sees the project but not the locked lines. Code, not a prompt, enforces it: 11K lines, 857 tests, MIT.
vps-forge: Agents With Root on a VPS Get Auto-Rollback, Self-Bans Drop From 7 in 60 Runs to 0
The open-source Agent Skill vps-forge gives agents a 5-minute dead man's switch. Across 120 benchmark runs, Opus 5.5 scores rose from 79–92 to 99–100, but each run costs three times as much.
BigSleep LLM Agent Finds 45 V8 Vulnerabilities, Fuzzilli 37, With 25% Overlap
Project Zero compared the BigSleep LLM agent and the Fuzzilli fuzzer on V8 over six months: 45 vulnerabilities vs 37. Overlap: a quarter in the tracker, about 60% in the experiment.
rwkv-quant: keeping one matrix in bf16 removes 39% of RWKV-7 2.9B's quantization loss
The rwkv-quant author found why COMPRESSION on RWKV-7 2.9B loses +4.04% perplexity versus +2.85% on 1.5B: one matrix is to blame, and the fix costs 0.5% of the file.
Qwen3-8B on the Intel Core Ultra 5 226V NPU: 10.7 tok/s at 15 W and 24 dB of Noise
Qwen3-8B on all three Lunar Lake blocks in a Ninkear L20 mini PC: the NPU hits 10.7 tok/s at 1.52 J/token, the GPU is twice as fast, and the machine stays quiet with CPU and GPU free.
Gemma 26B fine-tuned on a 24 GB MacBook: offline assistant gets 703 of 766 right
A developer fine-tuned Gemma 4 26B A4B with QLoRA on a laptop and moved 5 000 tokens of rules into the weights. Result: 703 of 766 conversations correct, median 1.64 s per reply.
Codex on GPT-6.1 Sol trims its window to 828 400 but still accepted an 888-thousand-token input
The config says model_context_window = 1 000 000, yet Codex reports 828 400. A 890 547-token test shows it is a history threshold, not a limit on a single input.
12 Pentesting Tools vs. Juice Shop: The Best AI Agent Found 5 Holes Out of 116
The author ran 6 AI agents, 4 DAST scanners and a Claude + Burp MCP combo against OWASP Juice Shop's 116 vulnerabilities. The hybrid won with 9/10; headline benchmark claims didn't hold up.
Replacing LLM weights with a recipe: how 16 GB of matrices are being folded into formulas
Instead of storing billions of numbers, a model can be described by a weight generator: hypernetworks, formulas, fractals. Habr examines what works and at what cost.
Habr Breaks Down Oracle Bias: AI Speeds Up Code, but Trust in It Isn't Growing
Part two of a series on reliable programming explains why process metrics rise after LLM adoption while trust in the code does not. It draws on data from DORA, Veracode and FormAI.
Geoffrey Huntley Sorts Dev Tools into S-Tier and F-Tier: What Decides It Is --help and Model Weights
Geoffrey Huntley: in the MCP vs. CLI debate, the winner is whoever made it into model weights. GitHub's gh CLI works without skills, and newcomers should start with --help.
Home-Built AI-Slop Detector Catches 5 of 12 Signs of AI Text With Code
An author mapped 12 signs of AI slop to his detector's checks: code catches five, two only partly. The other seven are left to an editor, and online detectors confuse humans with machines.
Claude Pushes an Amplitude Calculation from Eight Loops to Nine, Anthropic Says
Anthropic reports a nine-loop calculation by Claude; the previously published result for this amplitude stopped at eight loops. We look at how to verify scientific code written by AI.
Synthesized Voice Read a Novel in 1 h 58 min; the Surrounding Pipeline Code Was What Broke
The author of a Habr series voiced a 15 444-word novel: the voice held for two hours, while failures hit text splitting, phrase stitching and markup. A 14-phrase test set caught none.
Author on $200 Claude Plan Goes Back to 200K Context to Stay Within Limits
A $100 plan was enough in spring; by August the weekly limit ran out in 3 days. The developer blamed "everyday convenience" and went back to a 200K-token window.
Lambert: A Ban on Open Weights Would Pull Frontier Model APIs Down With It
Nathan Lambert of Interconnects examines the debate over cyber risks of open models: most documented attacks ran through closed systems, not open weights.
AI Reaches PLCs: Engineering Assistants Now Read Projects and Write Code
AI tools for PLC development are evolving from chatbots to full-fledged agents. PLC Copilot, Copia AI, and Siemens Eigen Engineering Agent now read projects, generate ST code, compile it, and fix errors.
2GIS's MCP Server: How to Stop LLMs From Making Up Addresses and Coordinates
2GIS has released an MCP server with six tools that connect AI agents directly to its geo-services: place search, geocoding, routes, isochrones, and static maps. The core problem: even with data access, models confidentl
October 2026 Video Model Rankings: Luma Ray 3.14 Wins on a Single Prompt for 5 Tokens
A test of 12 video models on an identical prompt (8 sec, 720p, single run): Luma Ray 3.14 scores 9.1 out of 10 while being the cheapest (5 tokens, 2-minute generation), while Seedance 2.5 is 27 times more expensive (138.
Why AI Agents Are Vulnerable to Prompt Injection — and How to Secure Them at the Architecture Level
Independent expert Andrey Biryukov explains the fundamental difference between AI systems and traditional software: in LLM agents, the control plane and data plane merge into one, making them vulnerable to injections. Se
AI Won't Replace Psychologists — But It Will Rebuild Their Workday by 2028
By 2028, artificial intelligence will be embedded in a psychologist's work not as a replacement, but as a tool for routine tasks. Help with journals, session prep, and formatting exercises — the system takes all of that
A Russian-Language Decision Model Trained on One Evening of GPU Time: What It Can Do and Where It Loses to Big LLMs
A developer on Habr fine-tuned the open Qwen3.5-4B model via LoRA to produce a specialized decision model for Russian. It picks from offered options in 60-100 ms with honest confidence calibration — its main edge over la
AI Agents: PRs Up 98%, But Review Time Slips 91%. Where the Queue Really Went
Developers using AI assistants close 98% more pull requests, but review time has grown 91%, and QA time 300%. AI sped up code generation, but review and integration throughput didn't keep pace. The bottleneck simply move
600 billion tokens later: how to manage AI agents decompiling a game
Three months of work by AI agents built on Claude, Codex, Luna, and other models — a full reconstruction of a C++ shooter. 99% of functions recovered, 83% match the original byte-for-byte. The key to success: byte-level
State of IT in 2026: Agents Wrote More Code Than Humans
In August 2026, agents opened more pull requests on GitHub than humans did. Since July, they've filed more tickets on Linear too. IDEs are disappearing, every company is building its own harness, and migrations now take
Russia Already Has 18 AI Code Agents, but 'Russian' Means Three Different Things
RuBench's census counted 18 Russian AI code agents (10 for IDE/CLI and 8 builders): from GigaCode, SourceCraft, and Kodik to T-Code, Koda, and 1C:Naparnike. We analyze the complete list, prices, and three architectures u
Eight Security Assumptions AI Agents Are Breaking: From Two Weeks to Ten Hours
AI agents have rewritten the economics of cyberattacks. Work that would take a human two weeks now takes agents ten hours. Here are eight critical security assumptions you need to drop right now.
Google teaches EmbeddingGemma 2 to understand text, images, video, and audio
Google DeepMind released the open multimodal embedding model EmbeddingGemma 2 on October 6. It converts text, code, images, video, and audio into vectors, placing them all in a unified 768-dimensional space. The first ve