OpenAI has reportedly paused work on its upcoming Astra model after internal evaluations found cybersecurity capabilities that could approach—possibly reach—the company’s “critical” threshold under its Preparedness Framework. The move signals that model providers are treating agentic cyber capability as a frontline safety risk, not just a technical benchmark. For defenders, it’s another reminder to plan for faster discovery-to-exploitation cycles as AI systems get more capable.
Source: Help Net Security
Metabase zero-day exploited to access Framework customer data
Framework notified customers that attackers exploited a Metabase zero-day to gain access to personal information including names, email and phone numbers, physical addresses, and login IP addresses. The breach appears to have avoided payment and order-record data, but the incident underscores how quickly analytics tools can become initial access paths. Teams should prioritize Metabase inventory, patch verification, and compensating controls (network restrictions and authentication hardening) around BI platforms.
Source: Help Net Security
N-able ships “Hotfix 2” for N-central after attackers keep exploiting CVE-2026-18577
N-able has released a second hotfix for its N-central RMM platform, emphasizing that Hotfix 2 is required even if Hotfix 1 was applied. The update reflects ongoing exploitation of CVE-2026-18577, along with additional hardening measures and indicators of compromise. For MSPs and enterprises, this is a clear operational cue: validate that the correct hotfix level is deployed and review RMM telemetry for suspicious activity.
Source: Help Net Security
Poland energy incident highlights a previously unseen pivot: private APNs into OT networks
Security reporting on a Polish combined heat and power (CHP) plant breach describes a novel entry vector: attackers moved from IT into OT using a private cellular network (private APN) set up by the local grid operator. CERT Polska described this as the first observed case of this pattern, where an “ordinary-looking” connectivity design becomes an attack route into PLC-connected environments. The takeaway for defenders: treat private connectivity (cellular, leased lines, remote management paths) as part of the OT threat model—segment, authenticate strongly, and monitor for anomalous control-plane behavior.
Source: Help Net Security
Metasploit opens its exploit engine to LLMs—“Ask AI, Get an Exploit” becomes easier
A reported Metasploit update (Metasploit 6.5) enables AI assistants to interact directly with the exploitation framework. While the intent is likely to accelerate security workflows, the same capability can lower friction for offensive experimentation and potentially accelerate misuse if access and safeguards are weak. Organizations should tighten governance around AI tooling (who can use it, where it connects, and what it can execute) and monitor for automated exploit-attempt patterns.
Source: InfosecNewsPaper
Anthropic is moving Claude Code “auto mode” toward default—faster action, higher stakes
Anthropic plans to make Claude Code’s auto mode the default for new sessions on Pro, Max, and Team plans, starting August 14, with an optional one-time prompt for users. In a controlled experiment with 1,053 testers, human review caught only 13.6% of dangerous commands, while auto mode caught 89%, implying stronger automated filtering of harmful actions. Still, defaulting to automation changes risk management: teams should revisit policies, approvals, and “what actions are allowed” for AI coding tools.
Source: Help Net Security
Edge is dropping older extensions after Manifest V2 retirement—privacy tools may break
Microsoft is retiring Manifest V2, the basis for older Edge extensions, which can cause some popular privacy tools to lose features or stop working. While not a vulnerability, the change can indirectly affect security posture by disrupting user protections and workflows used to block trackers or scripts. Security teams supporting end-user environments should inventory critical extensions and validate replacements before the deprecation impacts rollout timelines.
Source: MalwareBytes Blog
You May Also Be Interested In... CISA urges immediate patching of exploited Progress LoadMaster vulnerability
GitHub Dependabot malware alerts now cover eight ecosystems
Project CAV3RN continues: Google Apps Script C2 relay and DNS routing