Loading story…
ChatGPT and other AI-chatbot conversations are increasingly being introduced as evidence in U.S. criminal and civil cases because they generally lack attorney-client, medical, or therapist privilege. The Washington Post identified chatbot records in at least 12 public cases over two years; investigators and litigants may obtain them through consensual device searches, subpoenas, civil discovery, or provider disclosures in credible imminent-threat situations.
In litigation over the 2020 Watson Grinding explosion in Houston, plaintiffs obtained about 350 pages of 3M expert witness Josh Autenrieth's ChatGPT history after he used the tool to prepare an expert report. The logs reportedly included prompts seeking to frame 3M as not at fault; a jury awarded $61 million and assigned 30% fault to 3M. Organizations and legal teams should treat AI-chat records—especially those used to develop expert opinions or containing confidential material—as potentially discoverable records, while recognizing that European discovery rules and GDPR constraints may limit comparable access.
Track how attackers are adapting to this technology.
19 events from the most recent confirmed update back to the earliest known activity.
OpenAI disclosed data involving more than 80 accounts in response to government and law-enforcement requests, more than four times the number for the corresponding prior-year period.
R.K.C. used ChatGPT to ask about his father's comment concerning a potential settlement in his social-media litigation.
Teenager R.K.C. sued Meta, Snap, TikTok, and YouTube, alleging their platforms caused social-media addiction and mental-health harm.
Watson Grinding and Manufacturing in northwest Houston, Texas, exploded after propylene gas reportedly leaked from a worn, poorly crimped welding hose and accumulated in the building. The blast killed three people, injured dozens, and damaged about 200 homes and businesses.
OpenAI described safeguards intended to prevent ChatGPT from facilitating violence, including automated monitoring, contextual human review, account bans, and escalation of conversations presenting an imminent and credible risk of harm to law enforcement. It also described support measures for users in distress, including localized crisis resources and parental controls for teen accounts.
In The New York Times copyright lawsuit against OpenAI, a judge ordered OpenAI to retain ChatGPT conversation logs, including chats that users had requested be erased. OpenAI said the order required it to forgo privacy protections and retain data it had agreed to delete under EU and California privacy laws.
Bradley Heppner was convicted of securities fraud, wire fraud, and other charges.
A federal judge in New York ruled that financial executive Bradley Heppner's Claude conversations about possible fraud defenses were not privileged because Claude is not a lawyer and his attorneys had not directed him to seek chatbot advice.
In a Michigan employment dispute, a former employer argued that a tyre salesman's ChatGPT questions about recovering deleted Yahoo emails indicated an attempt to conceal evidence. A judge allowed the case to proceed and ordered the salesman to pay unspecified legal fees.
After Ryan Schaefer was accused of damaging 17 cars in a Missouri State University parking lot, police searched his phone with his consent and found ChatGPT messages sent shortly after the alleged vandalism.
Darren Zhou pleaded guilty and was sentenced to eight years of probation in the stalking and electronic-threat case.
Police alleged that Darren Zhou sent anonymous messages to his former girlfriend after their relationship ended, and he was arrested on stalking and electronic-threat charges.
OpenAI alerted the FBI after a ChatGPT user repeatedly described plans to harm an ex-girlfriend. The FBI shared the information with local Palm Beach County police, who identified the potential victim.
R.K.C. settled with Snap, TikTok, and YouTube and withdrew his claim against Meta.
Defense lawyers obtained R.K.C.'s ChatGPT usage history during discovery in his social-media addiction case. Morgan & Morgan partner Mike Morgan said the messages did not affect the outcome.
A jury awarded $61 million and assigned 30% of the fault for the Watson Grinding explosion to 3M, representing approximately $18.3 million of the verdict.
A judge allowed plaintiffs to obtain Autenrieth's ChatGPT prompts and conversations underlying his expert report. During a paused deposition, roughly 350 pages of chat history were retrieved; some chats reportedly had publicly accessible share links.
3M expert witness Josh Autenrieth used ChatGPT to prepare an opinion defending 3M's standard of care in the Watson Grinding litigation. He provided the model with hundreds of pages of court records, and the article reports that 85% to 90% of the final report came from ChatGPT output.
Residents and victims' families sued Watson Grinding and 3M, alleging that 3M failed to properly service the gas-detection system it was contracted to maintain.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Follow how adversaries are adapting to this technology, and where it touches your stack today.
6 references tracked. Mallory keeps watching after this page renders.
malware.news
Open sourcemalwarebytes.com
Open sourceosintteam.blog
Open sourceghacks.net
Open sourcewashingtonpost.com
Open sourceopenai.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
A U.S. court overseeing *The New York Times* copyright litigation ordered OpenAI to preserve ChatGPT output logs, including conversations users delete, as potential evidence. The preservation requirement creates a significant retention obligation for AI-service data that would ordinarily be subject to user deletion and minimization practices. The order raises compliance concerns under the EU General Data Protection Regulation (**GDPR**) and the California Consumer Privacy Act (**CCPA**), which establish rights to deletion and impose limits on retaining personal data. Organizations using or building generative-AI services should review litigation-hold processes, data-retention disclosures, cross-border data handling, and procedures for responding to privacy-rights requests when records must be retained for legal claims.
Sep 2, 2026
OpenAI has implemented significant changes to its data retention policies for ChatGPT users following a legal dispute with major news organizations, including The New York Times. The lawsuit alleged that users were leveraging ChatGPT to bypass paywalls, often by using temporary or deleted chat logs, prompting a court order that required OpenAI to preserve these logs indefinitely. OpenAI initially resisted the order, citing user privacy concerns, but ultimately lost the legal battle. By July, news plaintiffs began reviewing the preserved logs, which contained only ChatGPT's outputs, while attempts by some users to intervene in the case were denied due to their status as non-parties. On September 26, OpenAI was permitted to cease the controversial practice of retaining all output log data that would otherwise be deleted, following a joint motion approved by US Magistrate Judge Ona Wang. However, some users' deleted and temporary chats remain subject to monitoring under the agreement. This development marks a shift in OpenAI's approach to user data privacy and legal compliance, as the company balances regulatory demands with user expectations. Concurrently, OpenAI has experienced a dramatic increase in AI bot traffic, particularly after the release of its GPT-5 model on August 5, 2025. According to Akamai, OpenAI's bots—including ChatGPT-User, GPTBot, and OAI-SearchBot—now account for the majority of AI bot traffic across major web applications and APIs, with overall traffic rising nearly 300% year-to-date. The release of GPT-5 led to a surge in activity: OAI-SearchBot traffic increased by 136%, ChatGPT-User by 65%, and GPTBot by 66% during August and September. These changes in bot behavior are attributed both to increased adoption of the improved GPT-5 model and to modifications in how the bots interact with external sources. The period following the legal settlement and the GPT-5 launch has been characterized by unusual and intensified bot activity, impacting organizations that rely on web and API security. OpenAI's evolving platform and policy landscape underscore the complex interplay between technological advancement, user privacy, and legal oversight. Organizations are advised to monitor these developments closely, as changes in OpenAI's data handling and bot operations may have direct implications for data privacy, compliance, and web infrastructure security. The ongoing adjustments to OpenAI's practices reflect broader industry trends in AI governance and the challenges of managing large-scale AI deployments in a rapidly changing regulatory environment. Security teams should remain vigilant for further changes in bot behavior and data retention practices as OpenAI continues to adapt to legal and market pressures. The intersection of legal action, user privacy, and technical innovation at OpenAI provides a case study in the evolving risks and responsibilities associated with AI-driven platforms.
Jun 29, 2026
An investigation found that **264,078 shared conversations** across six AI chatbot platforms — **ChatGPT, Claude, Gemini, Grok, Copilot, and DeepSeek** — were publicly accessible and indexed by Archive.org after users generated web pages through built-in “Share” features. The report said **96,477** of the exposed conversations were from 2026 alone and that the issue was not caused by a platform breach, but by publicly reachable links created for sharing. Reportedly exposed material included names, phone numbers, home addresses, national ID numbers, immigration and residency records, administrative documents, retirement letters, and intimate personal exchanges. The disclosures renewed scrutiny of privacy controls on major AI services, particularly after shared **Claude** conversations were reportedly discoverable through Google indexing, leading Anthropic to disable that indexing path. Guidance published afterward urged users of **ChatGPT, Gemini, Copilot, and Claude** to reduce exposure by revoking public share links, using temporary or incognito chats, deleting chat histories, disabling model-training use where available, and turning off memory or personalization settings that can retain additional context or connected-service data.
Aug 4, 2026