Dev tools ethics flare up

GitHub posted an update on March 25, 2026 saying Copilot interaction data from Copilot Free, Pro and Pro+ users will be used to train its models by default starting April 24, 2026 unless those users opt out. (github.blog) GitHub’s public description of “interaction data” lists inputs and outputs, accepted or modified code snippets, surrounding cursor context, file and directory names, navigation patterns, Copilot chat logs and thumbs‑up/thumbs‑down feedback as items that can feed model training. (github.blog) The company maintains that private repository source code “at rest” will not be used for training, but admitted that content captured while a user is actively using Copilot inside a private repo can be collected as interaction data, a distinction critics say creates a technical loophole. (github.blog; theregister.com) Copilot Business and Copilot Enterprise customers are explicitly excluded from the default training change, while individual Free/Pro/Pro+ accounts are opted in by default — a tiered protection model the developer community has described as effectively “pay for privacy.” (github.blog; techspot.com) Anthropic’s Claude uses a different approach: consumer users are presented with a data‑training preference and had to choose by September 28, 2025 whether chats and coding sessions may be used to train models, with allowed training data retention described as up to five years. (code.claude.com; techcrunch.com) Security researchers and community posts have pointed to prior incidents where Copilot surfaced content from repositories that were public at indexing time but later made private, and outlets published step‑by‑step opt‑out guides after GitHub’s announcement because users were given roughly a 30‑day window before the April 24 change took effect. (techcrunch.com; computeleap.com)