Fake Weights & Biases / Comet ML MLops experiment tracking subscription payment failed, experiment runs suspended, model training logs no longer captured, or hyperparameter sweeps disabled phishing

Phishing emails impersonating Weights & Biases (W&B) or Comet ML claiming the MLops subscription payment has failed, experiment runs are suspended, model training logs are no longer being captured, hyperparameter sweeps are disabled, or model artifacts are no longer accessible — directing them to update billing or restore access through a credential-harvesting portal. A distinct attack category targeting ML experiment tracking platforms whose workspace is the central hub for all model training runs, results, and versioned artifacts during active ML development cycles. Key facts: (1) Weights & Biases serves 50,000+ ML practitioners and 1,000+ enterprise teams ($0/developer, $50/month Team, enterprise pricing) including teams at OpenAI, Toyota Research, and 90% of the top AI research labs — W&B is the de-facto standard experiment tracking tool for deep learning; every training run (whether on a single GPU or a 1,000-GPU cluster) streams metrics, loss curves, hyperparameters, system resources, and model outputs to W&B in real time; a W&B Team subscription suspension stops all experiment logging immediately — runs still execute on the compute hardware, but none of the metrics, gradients, loss curves, or evaluation results are captured; (2) The 'model training logs are no longer being captured' hook is exceptionally high-urgency for ML teams: a hyperparameter sweep configured to run 500 experiments across a cluster may cost $10,000-$50,000 in compute time — if W&B stops logging during the sweep, the team pays the full compute cost but loses all results; every GPU-hour spent during the suspension produces no usable data; ML experiments are inherently non-reproducible in real-world settings (random seeds, data shuffling, hardware variation), so lost logs cannot simply be re-run; (3) Comet ML serves 100,000+ ML practitioners (free + paid enterprise) with experiment tracking, model monitoring, and LLM evaluation capabilities — Comet ML is used by data science teams at Amazon, Uber, and Samsung for tracking model training across frameworks including TensorFlow, PyTorch, scikit-learn, and XGBoost; a Comet account suspension simultaneously loses all experiment history, model comparisons, and training run metadata that the team uses to decide which model version to deploy; (4) W&B and Comet ML are embedded in training scripts via a 2-line SDK integration (`wandb.init()`, `wandb.log()`), meaning suspension doesn't require any code change to cause harm — it's completely invisible to the training loop until the team checks the dashboard and finds empty runs; ML teams running weekend GPU jobs often don't notice until Monday morning, by which point 48+ hours of compute has produced no logged data; (5) W&B and Comet ML credentials expose the complete ML research and model development portfolio: every model architecture tried, every hyperparameter configuration tested, every loss curve showing what works and what doesn't — proprietary ML research that represents the core competitive IP of any AI-first company, plus the artifact registry containing trained model weights. Warning signs: sender not wandb.ai or comet.ml; genuine W&B billing at wandb.ai/billing; Comet ML billing at www.comet.com/billing.