ChatGPT Share Link Chain Exposed Accounts; MathLive Changes Origin Checks

·2 views

Security firm Parameter reported a chain of flaws that let an attacker run JavaScript as chatgpt.com in a victim's signed-in session by sending a shared conversation link. The chain began with a postMessage listener in the open-source MathLive library that accepted messages from any origin, a default documented in its API reference. Parameter said the flaw allowed reading a victim's private conversations, though it reported no real-world exploitation and tested only between its own accounts. OpenAI fixed the ChatGPT side on October 26, and MathLive released version 0.111.2 on October 31 to validate origins by default. The MathLive changelog does not mention ChatGPT, OpenAI, Parameter, or a CVE, and it is unclear whether the report prompted the change.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why None Is Not Enough: Sentinel Values for PATCH Updates in Python

A recent Hacker News discussion on sentinel values in Python prompted the author to recall a production bug from an update function that used None to mean both "not provided" and "clear the value." The problem surfaces clearly in PATCH APIs, where a field can be absent, explicitly null, or set to a value, and a plain None default cannot distinguish the first two cases. The article reviews the common object()-based sentinel and points out its drawbacks: an unhelpful repr, unreliable copying and pickling, and poor type-checker support. It then proposes an Enum-based sentinel that keeps a readable repr, survives deepcopy and pickle as a singleton, and lets type checkers narrow the type correctly.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Student Team Builds Offline AI Tool to Log Developer Decisions at Overnight Hackathon

A student developer describes their first overnight hackathon at Hacktoberfest Hack Day, held on the NST campus in Sonipat. Their four-person team built IssueFlowP6, an offline command-line tool that extracts architectural decisions from developer chat logs using a local Qwen 2.5 model via Ollama. The tool is designed to avoid hallucinations and data leaks associated with cloud-based models by verifying extracted quotes with a deterministic engine. The author recounts a Git mishap during the sprint that was resolved by isolating their work on a separate branch. The team finished a working pipeline before the deadline and submitted the write-up to a Hacktoberfest blog challenge.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Sipmap team builds DejaView to recover rejected ideas and reasons from old open source discussions

At a Hacktoberfest Hack Day in Sonipat, a four-person team took on a problem asking for a system that reads past discussions and recovers the decisions made in them. They were inspired by the recurring loss of reasoning in open source projects, citing Python's rejected 'x as y' proposal as an example where the rationale lived only in a heated thread. Their tool, DejaView, reads a Python PEP discussion and returns the rejected ideas, why each failed, and the post where it was said. To work within the twelve-hour limit, the members divided the work by area and agreed on folder ownership and interfaces up front. Early on, they hit an API quirk where a thread request returned success with no posts, because the service uses separate global and human-visible post IDs.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AIClaw 3.7.29 Separates Agent Progress From Final Answers

AIClaw 3.7.29 changes its desktop conversation view so that execution progress is kept apart from the final answer. Progress messages and tool activity stay in chronological order within each turn, rather than being merged into one prose block. The final answer appears with its own presentation, copy action, and file references, and is shown only once the turn is no longer running. Tool-created files such as helper scripts or screenshots are no longer treated automatically as deliverables. The article notes that these are display rules and do not verify whether the task itself succeeded.

0 comments Read more at DEV Community