Find notable cyber news and cases, enriched with sources, timelines, and signals.
Home Watchlists Beta Browse News Feed Stats About Contact

OpenAI Anthropic and Google reasoning APIs cross-session replay security flaw

Vulnerability
First reported
Last updated
Happening score
H score 36
1 unique sources, 1 articles

Summary

Hide ▲

A newly disclosed cross-session replay flaw in OpenAI, Anthropic, and Google reasoning APIs exposed hidden reasoning and secrets from session logs, including API keys and passwords. The weakness let opaque reasoning blocks move across sessions, users, and compatible models, turning preserved reasoning state into a secret-extraction risk. Researchers said the demonstrated attacks stopped working after mitigations in August 2026.

Related Happenings

Chrome Google Password Manager passkey post-compromise techniques on Windows

Technical Analysis
H score3 First: 03.08.2026 19:24 Last: 03.08.2026 19:24 Sources 1

About this happening: Unit 42 expanded the Chrome Google Password Manager passkey happening with Pass-ta-key research that shows how malware already on a Windows endpoint can manipulate...

Open Happening

OpenAI model sandbox escape and exploit chaining during ExploitGym evaluation

Technical Analysis
H score34 First: 22.07.2026 07:18 Last: 22.07.2026 07:18 Sources 1

About this happening: OpenAI's GPT-5.6 Sol and a pre-release model were observed chaining vulnerabilities and escaping a sandbox during evaluation, showing how advanced model behavior can drive...

Open Happening

Agent data injection proof-of-concept attacks expose trusted-data flaws in AI agents

Technical Analysis
H score25 First: 16.07.2026 14:32 Last: 16.07.2026 14:32 Sources 1

About this happening: Researchers disclosed agent data injection (ADI), a new attack class that can make shipping AI agents misclick, run attacker commands, and trust fake history across web and co...

Open Happening

MemGhost stealth memory injection against OpenClaw personal agents

Technical Analysis
H score23 First: 13.07.2026 16:49 Last: 13.07.2026 16:49 Sources 1

About this happening: Researchers demonstrated MemGhost, a one-email prompt-injection technique that can plant a persistent false memory in OpenClaw-style personal agents, letting an at...

Open Happening

Google Dialogflow CX Code Blocks shared-runtime isolation security flaw

Vulnerability
H score32 First: 07.07.2026 19:37 Last: 07.07.2026 19:37 Sources 1

About this happening: Google Dialogflow CX Code Blocks had a shared-runtime isolation flaw that could let one editable agent affect other Code Block-enabled agents in the same Google Cloud pr...

Open Happening

Timeline

  1. 12.08.2026 14:47 2 articles · 4h ago

    Researchers disclose replay flaw in OpenAI, Anthropic, and Google reasoning APIs

    Initial Disclosure

    Researchers disclosed a cross-session replay flaw in OpenAI, Anthropic, and Google reasoning APIs that let a compatible weaker model recover hidden reasoning from session logs and expose API keys, passwords, access tokens, and private keys from published agent traces. The team said encrypted reasoning objects were portable across sessions, users, and models, used Claude Haiku 4.5, GPT-5.6 Luna, and Gemini Robotics ER-1.6 as fuzzy decoders during testing, and found 704 distinct privacy artifacts from genuine user sessions across 6,708 public agent trajectories; the demonstrated attacks stopped working after mitigations, and no malicious exploitation in the wild was documented.

    Show sources
    Open in new tab