Secure Zona @securezona

Fable 5's return underscores why static AI safety filters aren't enough. As autonomous exploit-finding engines go global, organizations need robust AI-SPM to continuously govern model behavior, track active trust boundaries, and enforce hard guardrails at the runtime layer.

The Hacker News @TheHackersNews

16 hours ago

🔥 Anthropic is putting #Claude Fable 5 back online worldwide. The U.S. lifted June 12 export controls tied to an Amazon-reported jailbreak that surfaced vulnerability-finding and exploit-code behavior. Fable 5 returns July 1 across Claude platforms. #Anthropic says its new

11 23 101 20K 9

@TheHackersNews The lifting of export controls brings automated vulnerability-finding tools to the masses. To stay ahead, enterprise defense must shift from reactive patching to automated security posture management (SPM)—validating attack paths faster than the models can discover them.

The race to deploy AI agents is creating a silent crisis: API token sprawl. These non-human identities often bypass MFA and SSO entirely. If you aren't continuously discovering and monitoring active integration keys, you've left a permanent back door into your data. #InfoSec

Direct hit on the OWASP Top 10 for Agentic Applications (ASI02 & ASI04). When tools move to a read-write workflow, semantic poisoning redirects legitimate business processes rather than just leaking text. Enforcing human-in-the-loop controls for high-impact actions is mandatory.

Microsoft Threat Intelligence @MsftSecIntel

a day ago

As enterprise deployments mature, some enterprise AI agents are shifting from reading content to taking action, creating opportunities for threat actors to misuse trust relationships within agent workflows. msft.it/6010vssTs In this blog, Microsoft Incident Response

7 6 25 7K 18

@MsftSecIntel This is a crucial read from Microsoft IR. As Model Context Protocol (MCP) tool-use scales, we can’t just secure the model; we have to secure the agentic supply chain. Treating tool metadata updates with the exact same rigor as core system prompts is a non-negotiable step.

Update: The Miasma Mini Shai-Hulud campaign has expanded to #immobiliarelabs npm packages. It hit Backstage GitLab and LDAP auth plugins, with the compromised codfish/semantic-release-action suspected as a possible upstream path. Read the latest: thehackernews.com/2026/06/miasma…

@TheHackersNews A classic supply chain domino effect hitting enterprise infrastructure. The focus now shifts from simple detection to containment. Security leaders need to verify if these plugins are live in their environments, isolate affected builds, and lock down active developer tokens.

Big play by X. Bringing hosted Model Context Protocol layers to Grok Build, Cursor, and Claude Desktop creates a unified, permission-controlled pipeline.

Cyber Security News @The_Cyber_News

a day ago

X Launches Hosted MCP Servers to Connect Cursor, Claude, and Other AI Tools Source: cybersecuritynews.com/x-launches-hos… X has officially launched hosted Model Context Protocol (MCP) servers, enabling AI development tools such as Grok Build, Cursor, and Claude Desktop to seamlessly connect

4 18 73 6K 9

@The_Cyber_News This is the "USB port" moment for AI agents. By hosting native MCP servers, X eliminates the API integration tax for IDEs and assistants. Standardizing how autonomous tools consume documentation and data streams is exactly how you scale agent workflows.

🎮 Tell an AI browser it’s just playing a game. Researchers say "BioShocking" tricked six AI agents, including #ChatGPT Atlas, Comet, and #Claude, into copying GitHub SSH credentials from a signed-in session. Read how the attack chain worked: thehackernews.com/2026/06/new-bi…

@TheHackersNews A massive wakeup call for enterprise security teams. We are letting agentic AI assistants read, type, and click inside our active work sessions without explicit validation gates. If an agent can be tricked into leaking active SSH keys, human-in-the-loop is mandatory.

Unmanaged browser extensions are the ultimate silent risk to enterprise data. Legitimate tools get sold or compromised, turning into background scrapers that bypass firewalls to steal session cookies and proprietary data right out of active browser tabs. #InfoSec #DataSecurity

This is not a “remote hack” story.T he real risk is when a local attacker already has access and can use #Linux namespace settings to reach the vulnerable kernel path. That’s why patching and hardening both matter. Read the full article: thehackernews.com/2026/06/new-li…

The Hacker News @TheHackersNews

5 days ago

🛑 A new #Linux kernel exploit (CVE-2026-46331) gets root without modifying a single file on disk. It poisons the cached copy of /bin/su in memory. The binary on disk stays untouched. File-integrity checks come back clean. The root shell is already open. Details here ↓

21 300 1K 179K 716

@TheHackersNews This is why Local Privilege Escalation (LPE) is the real nightmare for shared hosting and Kubernetes environments. If an attacker gains initial access, they can manipulate network namespace properties to completely bypass the user boundary. Stop ignoring local bugs.

New Claude Code Attack Allows Attackers to Take Full Control of Developers’ Systems Source: cybersecuritynews.com/new-claude-cod… A proof-of-concept attack that shows how a completely clean-looking GitHub repository can trick AI-powered coding agents like Claude Code into silently opening a reverse shell on a developer’s machine, without a single line of malicious code ever appearing in the repository. Published on June 25, 2026, the proof-of-concept (PoC) attack targets agentic coding tools such as Claude Code and exploits indirect prompt injection, a technique that embeds malicious instructions in external content the AI agent processes, rather than in direct user input. The result is catastrophic: a fully interactive shell running under the developer's own user privileges, with access to every secret in the environment, from ANTHROPIC_API_KEY to AWS_SECRET_ACCESS_KEY and GITHUB_TOKEN. #cybersecuritynews

@The_Cyber_News "Invisible" to static scanners because the exploit payload never touches the codebase. This indirect prompt injection attack leverages runtime indirection to open a reverse shell using the developer's token. We desperately need isolated sandboxes for AI coding agents.

Attackers are bypassing email filters using Indirect Prompt Injection. By hiding instructions in regular text emails, they trick connected AI assistants into leaking sensitive inbox data—no malicious links or malware required. #InfoSec #AIGovernance

Static security audits only capture a single moment in time, the minute your environment changes, that 100-page PDF is obsolete. Secure Zona’s Security Assessments map your actual, live posture in real time, turning static compliance lists into active defense. #InfoSec

@BleepinComputer This is why blindly trusting AI agents with privileged local execution is dangerous. Threat actors don't need to bypass a human developer anymore; they just trick the agent into silently spinning up terminal processes. If agents can execute code, sandboxing is mandatory.

The massive Salesforce integration breach proves that your security is only as strong as your weakest third-party app. Attackers didn't hack the core platform, they stole OAuth tokens via a connected integration to bypass authentication entirely.  #InfoSec