artificial intelligence | Breaking Cybersecurity News | The Hacker News

Adversa AI has disclosed an attack technique that it says can cause xAI's Grok chatbot to send a user's name, approximate location, subscription tier, and the prompts from the ongoing conversation to an attacker-controlled server after the user asks it to summarize an ordinary web page. The AI security company, which has codenamed the technique "Cryptographic Context Injection ," said the transfer completed without a confirmation step and with no visible warning in its proof-of-concept demonstration. There is no patch, no CVE identifier, and no user-facing workaround, and the writeup does not report any exploitation in the wild. Asked which build was tested, Adversa told The Hacker News the target was the Grok web chat at grok.com running Grok 4.5 Fast, and that the attack was reproduced once on August 19, 2026. The writeup gives no success rate. The company said it has attempted the attack 20 times since June with a 40% success rate, and that the failures cam...

OpenAI on Tuesday revealed that it paused reinforcement learning ( RL ) training for its latest artificial intelligence (AI) models for two weeks while it shored up additional defenses and increased the scope of its monitoring to avert another Hugging Face-like incident . "As models become more capable, the risks associated with developing and testing them internally also grow," the AI company said . "Our standards for monitoring, alignment, and security must stay ahead of those risks. We wanted to take the time necessary to meet those standards, so we temporarily slowed the pace of scaling." The company said its largest planned frontier RL run remains on hold for the time being as it conducts smaller-scale training and evaluations to evaluate model behavior, validate its safeguards, and establish more concrete evidence of alignment before moving to the next phase. To that end, OpenAI said it plans to strengthen safeguards across its development process, inc...

A previously unreported cyber espionage operation dubbed SilkParasite has been observed targeting government bodies in Central Asia. The intrusion set makes use of seven remote access tool (RAT) families, five of which have never been previously documented: DriveSilkRAT, CookiETagRAT, NomadRAT, GoginRAT, and NodeEdgeRAT. SilkParasite, first discovered in late 2025, is assessed to be a China-nexus threat cluster with medium confidence. "What makes SilkParasite interesting is the traces of AI-assisted development running through otherwise expert code, which is a different thing from AI-generated malware," Bitdefender Labs said in a technical report shared with The Hacker News. Unlike other operations that rely on AI-generated malware, SilkParasite's arsenal exhibits all hallmarks typically associated with professional espionage tooling that's developed by a team of human operators while AI is likely used to streamline the process. The Romanian cybersecurity v...

Most email defenses still do the job they did a decade ago. Scan the message, look for something malicious, block it. That worked when the danger sat in the payload, a bad link or an attachment. It stopped working when the danger moved into the message's intent, and it is failing now that the sender is no longer a person. From Bad Content to Bad Intent to AI on Both Sides Phishing 1.0 was bad content. Malicious links, infected attachments, spam. Secure email gateways were built for this. Scan the message, match the signature, drop the bad stuff. That era is largely handled. Phishing 2.0 is bad intent. Business email compromise, executive impersonation, fake invoices, wire fraud. There is no malicious payload to scan, only social engineering that reads as a normal request from a person you trust. Gateways are blind to it because there is nothing in the content to flag. Behavioral analysis is the only thing that catches it, which is why some of us have spent the better part ...

Two critical vulnerabilities impacting MLflow, an open-source artificial intelligence (AI) platform, and FUXA, an open-source, web-based SCADA / HMI software built for operational technology (OT) and industrial automation, are witnessing malicious scanning and exploitation efforts. According to independent reports from watchTowr and VulnCheck, the vulnerabilities in question are as follows - CVE-2026-64849 (CVSS score: 9.3) - An unauthenticated Server-Side Request Forgery (SSRF) vulnerability in MLflow that can allow an attacker who can reach the Tracking Server (mlflow server) to issue HTTP requests to arbitrary internal cloud metadata endpoints and extract sensitive data. (Affects versions < 3.15.0) CVE-2026-25895 (CVSS score: 9.5) - A missing authentication for a critical function and path traversal vulnerability in FUXA that can allow an unauthenticated, remote attacker to write arbitrary files to the server file system and achieve remote code execution. (Affects ve...

Security researchers at Anthropic and Switzerland's EPFL have demonstrated that self-propagating payloads can spread from one artificial intelligence (AI) agent to the next through the editable system prompt files that autonomous agent harnesses use to carry state between sessions. The work, released as a preprint on August 10, 2026, tests the technique in a simulated six-agent coding collaboration and in a chain of paired agents modeled on OpenClaw , the open-source autonomous assistant formerly known as Clawdbot and Moltbot . There is no evidence that the technique has spread successfully in the wild, and the same paper reports that a review of archived posts from Moltbook, the social network for AI agents, found no successful agent-to-agent propagation despite several attempts. A one-paragraph warning added to an agent's system prompt reduced spread to near zero across the payloads tested. Fifteen generations of adversarial optimization run against that warning on C...

Cybersecurity researchers at Wiz have disclosed a new GitHub Actions workflow injection vulnerability in Snowflake's public  snowflakedb/snowflake-connector-net repository that it said could be exploited through a crafted GitHub issue to execute commands in a workflow containing internal Jira credentials. The issue was present in  .github/workflows/jira_issue.yml , which ran when a public issue was opened and exposed JIRA_BASE_URL, JIRA_USER_EMAIL, and JIRA_API_TOKEN to the same workflow step. The weakness was confined to the repository's CI/CD automation, with no affected Snowflake Connector for .NET release identified. The workflow inserted attacker-controlled issue title and body values directly into a shell run: block. It also checked github.event.pull_request.user.login even though the event was an issue, meaning the referenced pull request property did not exist. GitHub says , "If you attempt to dereference a nonexistent pr...

Cybersecurity researchers have traced the continued evolution of the Cavern (aka Cav3rn) command-and-control (C2) framework used by Iranian nation-state hackers in attacks targeting entities in Israel. Russian cybersecurity company Kaspersky said its ongoing monitoring of the threat activity cluster since December 2025 has led to the discovery of previously unreported components that expand the toolkit's communication capabilities. "The main finding is a complex C2 module that uses DNS A-record responses to choose between direct HTTPS and a Google Apps Script relay for each transaction," Kaspersky said in an analysis. "The same DNS infrastructure can validate and replace the relay deployment ID, allowing the operator to rotate the Google channel." Cavern, first publicly documented by Check Point Research in early July 2026, consists of multiple moving parts, including an Agent and an assortment of modules, that work in tandem to enable mission-specific...

The expensive attacks are not always the clever ones. This week had plenty of proof. Exposed services got hit, old bugs found fresh use, browser sessions became attack paths, and supply-chain problems kept spreading farther than the original compromise. A lot of it came down to access that was already there and defenses that assumed nobody would look too closely. So, nothing magical. Just a lot of small openings turning into bigger problems. Here’s what stood out. ⚡ Threat of the Week Suspected China APT Behind Exploitation of New VMware Flaw — A suspected China-nexus APT is assessed to be behind the exploitation of a newly patched security flaw in VMware vCenter. The attacks involve the exploitation of CVE-2026-59310 (CVSS score: 9.8), a severe directory-traversal vulnerability in the VMware vCenter server that could be weaponized by a malicious actor to execute arbitrary code. In at least one compromised instance, the attacks led to the deployment of a backdoor and. a rever...

Some weeks have one big security story. Others bring many smaller updates that are easy to miss but still matter. This week has plenty of them, covering cloud services, AI tools, malware, data breaches, scams, and new attack methods. The latest ThreatsDay Bulletin puts all of these short updates in one place, so you can quickly catch up on what happened, what changed, and what security teams should know. The threats change every week. Subscribe, and we’ll alert you when each new ThreatsDay Bulletin is out.

Companies are used to thinking about attackers as outsiders trying to break in. North Korean IT workers flip that model. They apply for jobs, pass interviews, receive legitimate credentials, and can end up inside the same systems companies spend millions trying to protect. That risk is no longer theoretical. The FBI is now investigating a North Korean remote IT worker who reportedly worked for a U.S. federal agency. For CISOs, the priority is clear: spot the warning signs before a fraudulent hire becomes trusted access. When the Threat Gets Hired A recent joint investigation by Mauro Eldritch ( BCA LTD ), Heiner García ( NorthScan ), and ANY.RUN showed what this looks like from inside the operation. Researchers deliberately hired suspected DPRK developers linked to Lazarus Group and gave them what looked like ordinary virtual desktops. In reality, they were controlled ANY.RUN Sandboxes, capturing their activity in real time. The operation exposed forged identities, re...

A newly disclosed flaw in the way OpenAI, Anthropic, and Google carried hidden AI reasoning between API calls let researchers recover internal reasoning and secrets from session logs, including API keys and passwords. The weakness affected encrypted reasoning objects used by the providers' reasoning APIs, where a block created in one session could be replayed into another and, during testing, even handed to a weaker model in the same provider family to make it reveal the hidden content. The team behind the paper  Stealing Reasoning Traces from Proprietary LLM APIs  demonstrated four abuse paths: stealing proprietary reasoning for  model distillation , extracting private data from other users' published traces, recovering harmful content concealed behind a safe visible answer, and hiding prompt injections inside opaque reasoning blocks. Across 6,708 public agent trajectories, the team decoded 315,320 thinking blocks. After excluding benchmark sources, it counted 704...

Anyone sharing their screen on a Zoom call could have taken over the computers of everyone watching, and anyone watching could have taken over the presenter's. The flaw sat in the annotation tool, the feature that lets participants draw and type on a shared screen, and it asked nothing of the victim beyond being in the meeting. No click, no download, no prompt, and nothing on screen to show it had happened. The patches are not new. Client fixes shipped in June and July, roughly two months before the flaws were made public, and no exploitation has been reported as of publication. None of the three identifiers appear in CISA's Known Exploited Vulnerabilities catalog. The versions that close them: Zoom Workplace, all supported platforms, before 7.1.5 and 7.0.6 in their respective branches Zoom Workplace VDI Client for Windows, before 7.0.11 and 6.6.16 Zoom Rooms and Zoom Meeting SDK, all platforms, before 7.1.0, and before 7.1.5 for the third flaw The research ...

Security researchers found a way to enter Microsoft SharePoint servers as any user, including an administrator, with no valid account. A significant part of the work that found it was done through an AI agent. The flaw, tracked as  CVE-2026-55040  (CVSS 9.1), affects SharePoint Server Subscription Edition, SharePoint Server 2019, and SharePoint Server 2016. Microsoft's affected-product list covers only those three on-premises editions, and SharePoint Online is not among them. It lets a remote unauthenticated attacker assume a chosen user's identity. The attack has one prerequisite: the intruder has to know which account they want to become, either by its Active Directory security identifier (SID) or its user principal name (UPN), which is formatted like an email address. Rapid7 then chained the bypass to a separate remote code execution flaw and ran code on the server with no credentials. Microsoft and the firm disclosed that second flaw on August 11 as  CVE-2026...

OpenAI on Monday unveiled a new cybersecurity-focused model called GPT‑5.6‑Cyber that it said is focused on vulnerability research, penetration testing, and incident response. "Built on GPT‑5.6 Sol, it is trained to improve capabilities on several specialized cybersecurity tasks (e.g., finding zero-day vulnerabilities and developing exploit chains) and to reduce refusals for certain higher-risk, dual-use cyber tasks," OpenAI said . The artificial intelligence (AI) company said it's making GPT 5.6 Cyber available through Daybreak Red, a new tier that provides access to its purpose-trained cybersecurity models to other firms for authorized vulnerability research, exploit validation, and security testing. GPT-5.6-Cyber, a more cyber-permissive version of GPT-5.6 Sol, builds upon GPT‑5.5‑Cyber , which OpenAI released in June 2026. To measure the reduced rate of refusals provided by GPT‑5.6‑Cyber through Daybreak Red access, OpenAI said it created an internal evaluat...

Security researchers invented a cryptocurrency startup, advertised developer jobs, and hired three people they believe were North Korean operatives. Every virtual machine the company issued was recording. The onboarding paperwork is the part hiring teams can use. The first hire claimed to live in Pasadena, Texas, then sent a California driver's license and a New York bank account. The researchers said the image metadata showed it had been processed with Google Gemini. They also reported a SynthID watermark, the invisible marker Google embeds in images its AI tools create or edit. The second supplied a Texas license, a valid Social Security number, and a bank account in Kansas City. The third sent a New York license belonging to someone else, a genuine iPhone 15 photograph with the GPS coordinates stripped. A successful placement gives the operative a real employee account and real access to source code and internal systems. The July 31 joint alert says North Korean IT work...

AI is helping development teams produce far more code, far faster. But security teams still have to review vulnerabilities, manage dependencies, prioritize fixes, and control risk at human speed. When software output jumps 10 to 50 times, the problem is no longer just finding vulnerabilities. It is keeping security from becoming the bottleneck, or worse, losing control of what gets shipped. In our latest webinar with Chainguard experts, “ The True Cost of Building at Machine Speed ,” you can now watch how security teams can keep AI-driven development fast without letting risk scale with it. For years, application security followed a familiar cycle: developers wrote code, scanners found problems, security teams prioritized them, and engineers fixed what mattered most. AI puts that model under pressure. If teams can suddenly create many times more code, security can also end up with many more components, dependencies, findings, and fixes to manage. More scanning alone does no...

A lot of security problems still begin with someone doing a completely normal thing. Cloning a repo. Answering a call. Leaving a box exposed. Trusting the default. That pretty much covers the mood this week. Old bugs are back, supply chains are getting stranger, and some exploit paths are so short you wonder what was supposed to stop them in the first place. That’s only part of it. Here’s everything else that made the Monday recap. ⚡ Threat of the Week Anthropic's Model Attempts to Poison Open-Source Project — A new evaluation conducted by the U.K. AI Security Institute (AISI) found that AI models with access to the internet reached out into the real world to target individuals and organizations autonomously across 10 of the total of 122 runs. Of 19 such actions recorded, 17 originated from Anthropic's Mythos 5 and the remaining two involved OpenAI's GPT-5.6-Sol with cyber classifiers. In the most serious case, Anthropic's Claude Mythos 5 spent 34 hours trying...

North Korea's state hackers are no longer content to type prompts into public chatbots. One of the country's main espionage groups has begun running artificial intelligence (AI) offline on its own servers, connecting document-search tools to files in its possession, and collecting the software parts needed to build AI into its malware. South Korean security firm Genians says it uncovered the setup after months of tracking and log analysis on infrastructure tied to Kimsuky, a hacking unit under North Korea's Reconnaissance General Bureau. Genians found no evidence that the group had trained an AI model of its own, and the firm does not offer that as reassurance. It describes an actor in a "research and knowledge acquisition" stage, assembling and testing existing tools rather than making new models, with the apparent aim of folding AI through the operation, from writing malware to analyzing data. For an intelligence unit that has spent years phishing governm...

OpenAI has announced that it's pausing some "internal activities" involving its upcoming artificial intelligence (AI) model Astra after an internal evaluation found it had made significant advancements in agentic coding and cybersecurity. In response to the discovery, the AI upstart said it's implementing security controls for higher-capability models and associated activities, such as isolated testing environments, restricted network and tool access, enhanced model weight protections and encryption, additional monitoring and detection capabilities, and sandboxed execution. "We are pausing internal activities involving Astra that do not yet meet these strengthened security control requirements," it said in a statement. "We have implemented universal monitoring for risky actions and misalignment across all agentic applications of Astra, including training and evaluation. Monitors evaluate the model's Chain of Thought and trigger a security re...