Loading story…
Security researchers and enterprise administrators are warning that connecting AI assistants such as Claude, ChatGPT, and Copilot to business platforms like Microsoft 365, Gmail, Slack, Dropbox, Zoom, Outlook, SharePoint, and OneDrive can sharply expand the attack surface. Reporting on the connector ecosystem said hundreds of integrations changed within weeks and many can pass data to additional AI services or subprocessors, complicating governance and creating hidden exposure outside primary enterprise controls. Anthropic documentation also notes that connected services process data on their own infrastructure and under their own terms, adding compliance concerns for organizations handling sensitive or regulated information.
The operational risk is not limited to data access: security coverage described "authority laundering", where untrusted input is transformed by one AI system into trusted instructions for another system authorized to act, including a cited case involving fund movement. Sysadmin discussions echoed those concerns, with organizations debating whether to allow broad OAuth access to Outlook, Teams, calendars, files, and even local device control, especially in HIPAA-regulated environments. Microsoft is meanwhile replacing Outlook's confusing Meeting Insights feature with a deeper Microsoft 365 Copilot meeting-prep experience, while vendors such as Ledger are promoting safer agent designs that keep AI actions read-only until a human approves sensitive operations on hardware.
Mallory correlates global threat intelligence with your attack surface — know if you’re exposed before adversaries strike.
6 events from the most recent confirmed update back to the earliest known activity.
PromptArmor found that 931 of 2,517 AI connectors changed over a six-week period and that 1,686 new tools were added to already-live connectors. The research also warned that connectors can route data to additional AI services and subprocessors, increasing hidden third-party exposure.
Microsoft confirmed that Outlook's Meeting Insights feature will be retired and replaced by a Microsoft 365 Copilot-based meeting preparation experience. The company said the replacement would begin rolling out in early September 2026.
Microsoft introduced the Outlook Meeting Insights feature in 2020. The feature used machine learning to surface files, emails, and other content relevant to upcoming meetings in the meeting invite interface.
Dark Reading described a recent attack in which Morse code text manipulated one AI agent into generating a legitimate-looking instruction for another AI system authorized to move funds. The article characterized the pattern as 'authority laundering' and warned against trusting AI outputs as authorized commands.
Ledger SAS launched Ledger Agent Stack, an open-source toolkit for AI agents that can read cryptocurrency balances and draft transactions while requiring human authorization on Ledger hardware before funds can move. The release also included supporting tools for developer and enterprise policy enforcement use cases.
Anthropic published Help Center documentation describing how to use connectors to extend Claude's capabilities. The documentation notes that connected services process data on their own infrastructure and under their own terms.
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
7 references tracked. Mallory keeps watching after this page renders.
theregister.com
Open sourcewindowslatest.com
Open sourcescworld.com
Open sourcedarkreading.com
Open sourcereddit.com
Open sourcereddit.com
Open sourcesupport.claude.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.
Security researchers reported that AI assistants with web-browsing/URL-fetch features can be repurposed as **stealthy command-and-control (C2) relays**, allowing attackers to proxy commands and exfiltrate data through legitimate-looking enterprise traffic. Check Point’s proof-of-concept, dubbed **“AI as a C2 proxy,”** demonstrated abuse paths against **Microsoft Copilot** and **xAI Grok**, including scenarios that do not require an API key or even a registered account, complicating common mitigations such as key revocation or account suspension. Separately, organizations are responding to broader **AI data security and privacy** concerns by restricting AI tooling: the **European Parliament** reportedly disabled built-in AI features on lawmakers’ work devices due to uncertainty over what data is uploaded to AI vendors’ cloud services and the risk of sensitive correspondence exposure. In healthcare, ECRI Institute researchers flagged **AI chatbots** as a leading 2026 health-tech hazard, citing safety, security, and privacy risks from unvalidated clinical use, alongside ongoing concerns about IT outages and legacy medical device cybersecurity—reinforcing that AI adoption is increasingly being treated as a governance and risk-management issue, not just a productivity upgrade.
Jun 29, 2026
Meta suspended its internal Model Capability Initiative after an internal error reportedly made sensitive employee monitoring data broadly accessible inside the company, intensifying criticism of a program that collected workplace activity from tools including Gmail, Google Chat, Meta AI, screenshots, and usage patterns. Meta said it found no evidence of malicious activity or an external breach, but the incident renewed concerns about consent, unencrypted storage, and how far companies are going to feed AI systems with employee data. Similar privacy concerns surfaced in the Netherlands, where a watchdog said the AIVD and MIVD intelligence services mishandled bulk datasets containing personal information through weak controls, excessive retention, and inadequate oversight, raising alarms that AI-assisted analysis could be applied to unlawfully sourced or improperly retained citizen data. At the same time, new reporting and vendor changes underscored how quickly enterprise AI is expanding the attack surface and governance burden. Orca Security said **81.2%** of organizations running AI packages had at least one known vulnerability and that **99.9%** of fixable AI vulnerability alerts remained unpatched, with risks spanning agent frameworks, vector databases, exposed API keys, excessive permissions, weak authentication, and internet-facing services such as Amazon SageMaker, Azure OpenAI, and Google Vertex AI. Microsoft is adding an in-meeting Teams control to disable Copilot, Facilitator, and recap features during sensitive discussions, while broader industry debate has focused on employees oversharing data with AI assistants, the need to govern agent identities and registries, and warnings that AI prompts, memory, and orchestration layers can leak or concentrate proprietary organizational knowledge if enterprises do not retain tight control over access, telemetry, and data handling.
Jul 16, 2026
Security leaders are warning that **AI agents are increasingly operating as “digital employees”** inside enterprise workflows—triaging alerts, coordinating investigations, and moving work across security tools—often with **broad permissions and limited governance**. The core risk highlighted is that organizations are deploying high-authority agents like plug-ins (reused service accounts, overbroad roles, weak oversight), creating fast-acting operators that can be manipulated and that lack the contextual judgment and policy awareness expected of human staff. Related commentary also raises concerns about **AI-to-AI communication** and “non-human-readable” behaviors that could reduce auditability and complicate investigations and control enforcement. In parallel, public examples show how quickly AI can accelerate **vulnerability discovery**: Microsoft Azure CTO Mark Russinovich reported using *Claude Opus 4.6* to decompile decades-old Apple II 6502 machine code and identify multiple issues, underscoring that similar techniques could be applied to **embedded/legacy firmware at scale**. Anthropic has also cautioned that advanced models can find high-severity flaws even in heavily tested codebases, reinforcing the likelihood that both defenders and attackers will leverage AI for faster bug-finding. Separate enterprise IT coverage notes that organizations are **reallocating budgets toward AI** by consolidating tools and renegotiating contracts, which can indirectly increase security exposure if cost-cutting reduces overlapping controls or if AI adoption outpaces governance and identity/access management maturity.
Jun 29, 2026