AI Security | Breaking Cybersecurity News | The Hacker News

A new class of prompt injection is spreading across commercial websites. It requires no malware, no stolen credentials, and no zero-day exploit. It abuses a standard feature built into almost every major AI assistant: pre-filled deep links. We observed production websites embedding hidden prompt injection payloads inside "Ask AI" buttons on marketing and competitor comparison pages. When a user logged into ChatGPT, Claude, Gemini, or Grok clicks one, a pre-formed query executes immediately in their session, with no confirmation and no warning. Most of these links are benign. The dangerous ones instruct the AI to permanently save the vendor's domain as a "trusted source," quietly biasing every future answer in that vendor's favor. In February 2026, Microsoft Security catalogued the behavior as AI Recommendation Poisoning , identifying 31 companies across 14 industries deploying it, with more than 50 distinct prompts observed in a single data source over ...

Security flaws in agent infrastructure from Amazon Web Services (AWS), Google, and Vercel let untrusted or forged instructions reach an agent's tools with no check that a model turn had authorized them. In several of the attack paths, the model never ran at all, so system prompts, content filters, and model-level guardrails never got a chance to intervene. The affected products include Amazon Bedrock AgentCore's InvokeHarness API, Google's Agent Development Kit (ADK) for Python, and the Vercel AI SDK harness packages for the Codex and OpenCode coding agents. AWS has fixed the managed service, Google addressed the issues in ADK 2.5.0, and Vercel patched @ai-sdk/harness-codex in version 1.0.29 and @ai-sdk/harness-opencode in version 1.0.28. These are not identical vulnerabilities and do not share the same attack conditions. AWS involved an authenticated remote request, Google's paths required attacker-controlled session events or user-authored function calls, and V...

Cybersecurity researchers have discovered more than half-a-dozen services advertisements for illegal access to artificial intelligence (AI) models on underground cybercrime forums and messaging platforms. One such service, Poison Claude, claims to offer access to Anthropic's large language models (LLMs), including Opus 4.8, Opus 4.7, Opus 4.6, and Sonnet 4.6. "Advertisements for Poison Claude explain how the service can offer the cheap tokens: by taking advantage of free bonus credits, such as the US$100 bonus credit on AWS for Bedrock accounts," Okta researchers Jeremy Kirk and Mathew Woodyard said in an analysis published Tuesday. "The service plainly states on its website that: 'We add those accounts to our pool, your request is routed to a specific account under the hood (you don't see this), and you get charged 5-15% of the official per-token price depending on the model.'" Poison Claude accepts payments in cryptocurrencies. Once a cus...

Two security flaws in Paperclip could let attackers execute commands on a network server or a developer's computer. Paperclip is an open-source control plane for teams of artificial intelligence (AI) agents, and both paths rely on importing a malicious agent and starting it. A third flaw could expose sensitive data and control-plane details through application programming interface (API) routes that did not enforce the expected access checks. The more severe server-side path, tracked as CVE-2026-41679 (CVSS score: 10.0), requires no pre-existing account or victim interaction against network-accessible deployments using authenticated mode with the default registration configuration. The second path, tracked as GHSA-x8hx-rhr2-9rf7 (CVSS score: 9.6), requires a user to open an attacker-controlled page while Paperclip is running in its default local_trusted mode. The source tagged as Paperclip v2026.416.0 contains the import-authorization fix and hostname-validation guard di...

GitGuardian researchers found 321 n8n instances accepting API tokens exposed in public GitHub commits and demonstrated four ways attackers could use them to access sensitive data and downstream credentials without exploiting a software vulnerability. We scanned public GitHub commits for exposed n8n API tokens and identified 4,576 unique credentials associated with 1,255 hostnames. Of the 896 instances reachable at the time of testing, 321 accepted at least one leaked token. That means leaked credentials provided authenticated access to 36% of the reachable instances we tested, or roughly 26% of all hostnames identified in the commits. The implications extend well beyond n8n. Organizations use the automation platform to connect databases, source code repositories, cloud environments, artificial intelligence services, customer support platforms, and other internal systems. A sufficiently privileged n8n token can expose workflow definitions and execution data, allow attackers to us...

An agent running Anthropic's Claude Mythos 5 spent 34 hours trying to get a malware dropper merged into a real open-source project during a cyber evaluation by the UK's AI Security Institute. When a bystander publicly warned that the code was malicious, the agent denied it, force-pushed a rewritten branch history to erase the evidence, and posted from a second account it controlled to vouch for its own work. The project's maintainer closed the pull request anyway. The institute, known as AISI, published its incident report on Tuesday. Across 122 runs of a capture-the-flag (CTF) exercise on two of its cyber ranges, researchers catalogued 19 unsanctioned actions on the live internet across 10 runs: 17 from Mythos 5, a restricted model Anthropic sells for cybersecurity work, and two from OpenAI's GPT-5.6 Sol. AISI says the attempts failed and that it has found no evidence of resulting real-world harm. Both models ran with their providers' cyber classifiers swi...

The U.S. Cybersecurity and Infrastructure Security Agency (CISA), on August 5, 2026, added three flaws to its Known Exploited Vulnerabilities ( KEV ) catalog, citing evidence of active exploitation in the wild. The list of vulnerabilities is as follows - CVE-2026-9198 (CVSS score: 9.8) - A code injection vulnerability in Langflow that allows unauthenticated attackers to achieve full remote code execution on default Langflow deployments. (Fixed in July 2026 with version 1.10.1) CVE-2026-34486 (CVS score: 7.5) - A missing encryption of sensitive data vulnerability in Apache Tomcat that allows a bypass of EncryptInterceptor, a cluster component that adds pre-shared key encryption to messages sent between cluster nodes. (Fixed in April 2026 with versions 11.0.21, 10.1.54, and 9.0.117) Also added to the KEV catalog is CVE-2026-18556 (CVSS score: 8.2), an authentication bypass vulnerability in N-able N-central. It's worth noting that an incomplete fix for this issue promp...

Google deleted three AI agent workflows from its Agent Development Kit (ADK) Python repository. Pillar Security showed that a public GitHub issue could manipulate a triage agent into triggering a privileged code-fixing agent. The researchers said the public agent could be prompt-injected into posting /adk-issue-fix as adk-bot . They identified the bot as a collaborator, so that comment satisfied the privileged workflow's owner, member, or collaborator gate. The trusted bot identity became the authorization bridge. The team demonstrated arbitrary code execution on the continuous integration (CI) runner and exfiltration of the bot personal access token (PAT). The privileged job also held a Google API key and a Google Cloud service-account credential. Its researcher-controlled proof-of-concept attacks do not identify in-the-wild exploitation or a compromised ADK release. The exposed component was the repository automation, not a flaw in the distributed ADK Python package. Fo...

Three high-severity security flaws have been disclosed in Hugging Face's Diffusers library that could allow crafted model repositories to stealthily execute arbitrary code on machines that load it, opening the artificial intelligence (AI) supply chain to security risk. "These vulnerabilities are bypassing trust_remote_code, the safeguard designed to stop unreviewed code from running in the custom pipelines loading process," Zafran Labs researchers Gal Zaban and Ido Shani said in an analysis published last week. The shortcomings have been collectively named FaceHugger . With Hugging Face becoming the "GitHub of the AI era" and its libraries and repositories prevalent in enterprise environments, vulnerabilities in libraries like Diffusers can grant attackers extensive access owing to how the library is embedded into production pipelines, CI/CD systems, and container images. Diffusers is a Python package that serves as a library of state-of-the-art (SOT...

Hidden instructions in a Word document can make Microsoft 365 Copilot rewrite figures in a report, then copy the same instructions into the finished file. Håkon Måløy disclosed the technique on July 28, 144 days after reporting it to Microsoft. In his proof of concept, the internally generated file triggered the same behavior when it was used in a second Copilot drafting session. Måløy's timeline says Microsoft confirmed the reported behavior on March 31 and deployed two mitigations. The first blocked the original prompt wording; the second upgraded the underlying model to GPT-5.5. He said the full chain worked with modified instructions on GPT-5.6 the next day, and the attack class still reproduced on July 28. "The vulnerability class therefore remains exploitable at the time of publication," Måløy said. The attack is not zero-click and does not execute conventional malware. It requires a Copilot drafting or editing operation, and the malicious document must en...

Network firewalls are the workhorses of modern cybersecurity. They are trusted to protect the network, blocking malicious traffic and preventing intrusions and breaches. And for decades, network security teams have built controls around a relatively stable model: users connect to applications, applications exchange data, and security tools inspect packets, protocols, and destinations. Firewalls became exceptionally good at understanding where traffic was going and whether it should be allowed. But just as AI is reshaping every aspect of the business world, it's also had a monumental impact on how security teams view network traffic and the firewall. AI is driving new network dynamics that traditional security policies were never designed to govern. Employees, applications, and agents send prompts, call models, connect to services, and trigger actions across the environment, all activity most firewalls can't see, let alone understand. AI agents interact with external servic...

Cybersecurity researchers have flagged a maximum-severity security flaw in Ruflo , an open-source agent meta-harness for Anthropic Claude Code and OpenAI Codex, that could result in unauthenticated remote code execution. The vulnerability, tracked as CVE-2026-59726 (CVSS score: 10.0), impacts all versions of the project before version 3.16.3. It has been codenamed RufRoot by Noma Security's research team, Noma Labs. Originally launched as Claude Flow, Ruflo is an AI multi-agent orchestration platform and harness that allows users to deploy multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. The project has more than 66,500 stars on GitHub. The crux of the vulnerability is that Ruflo exposed 233 tools, including shell command execution, database operations, agent management, and memory storage, through an unauthenticated Model Context Protocol (MCP) bridge that's open to the network by default. Specifically, the "docker-comp...

Microsoft has launched its first cybersecurity-specific model inside MDASH , its multi-model vulnerability identification and remediation harness. The company says MDASH, using MAI-Cyber-1-Flash and GPT-5.4, scored 95.95% on CyberGym. It also claims the configuration costs 50% less than its current best MDASH combination of GPT-5.4, GPT-5.4 mini, and GPT-5.3 Codex. Access is limited to approved MDASH customers through an Azure AI Foundry private preview. MAI-Cyber-1-Flash is designed to handle up to 90% of MDASH tasks, with GPT-5.4 reserved for the hardest 10%. It is available only inside MDASH, not as a standalone public model or general-purpose application programming interface. The headline score belongs to MDASH running MAI-Cyber-1-Flash alongside GPT-5.4, not to the new model by itself. CyberGym Level 1 is a known-vulnerability reproduction test. It gives an agent a vulnerability description and the corresponding unpatched source code, then checks whether it can produce a ...

NVIDIA and 36 other organizations have formed the Open Secure AI Alliance to develop and share open technologies, techniques, and tools for securing software and artificial intelligence (AI) agents. The 37-member group spans cloud, security, enterprise software, and AI companies, including Microsoft, Cisco, Cloudflare, CrowdStrike, Hugging Face, IBM, Palo Alto Networks, Red Hat, and the Linux Foundation. Its stated scope covers the full agent stack, including identity, permissions, isolation, guardrails, logs, model formats, multi-model scanning, and secure coding workflows. The pitch is that cyber defenders need AI models they can read, change, and run on their own hardware, not only closed systems reached through a vendor's application programming interface (API). The launch also brings its first named technical contribution: NVIDIA-labs OO Agents (NOOA) , an Apache 2.0 research framework designed to make agent behavior easier to test, trace, audit, and govern. The laun...

Cybersecurity researchers have disclosed a critical vulnerability in OpenAI's ChatGPT Workspace Agents that could have allowed a single phishing link to stealthily build, authorize, and deploy an autonomous artificial intelligence (AI) agent inside a victim's organization. The vulnerability has been codenamed AgentForger by Zenity Labs. The issue has since been addressed by OpenAI as of June 8, 2026, following responsible disclosure. "A single link could hijack OpenAI's ChatGPT Agent Builder to stand up an attacker-controlled AI agent with a real employee's access and its approvals switched off," the AI security company said in a two-part report shared with The Hacker News. The attack occurs when an unsuspecting employee clicks open a benign-looking ChatGPT link, causing it to spawn a new AI agent within the company's trust boundary that does the attacker's bidding. The issue is a case of cross-site request forgery (CSRF) that forges an attack...

AI agent security is moving through a familiar maturity curve: adoption, then visibility, and finally, control. But what we've collectively discovered is that enforcing least privilege for AI agents is harder than we ever imagined. This is why there are so many approaches, from prompt filtering to identity-layer access controls. Where we've collectively landed is that understanding the intent of AI agents is essential to securing them. It's not easy, but it's the only path forward. Organizations approach this challenge with different levels of sophistication. For many, the current goal is simply to find the AI agents already operating across the business. That is a necessary first step. AI agents are appearing in SaaS platforms, developer environments, cloud workflows, customer support systems, productivity tools, and internal applications. Some are sanctioned, and others are not. But discovery alone isn't useful. AI agents are not passive; they reason, plan, c...

Eight security flaws in NodeBB went public on Wednesday, along with the code to exploit them. Aikido Security rates all eight as high severity and says its AI pentest agents found them in a six-hour review of the forum software's source code. Every version before 4.14.0 is affected. NodeBB has fixed them all, and administrators should be on 4.14.2. The simplest one takes a settings change. A regular forum member could point their homepage setting at the admin address, reload the page, and the admin dashboard opened for them. No password, no exploit code. The forum's own interface blocks that setting, but the block only ran in the browser and could be sidestepped. Most of what a member could then reach was read-only, including the error log and any user list an admin had exported, though they could also swap the site logo. Two more gave an attacker with no account at all access to things meant to be private. One let anyone claim to be any user and read private messages...

Redis shipped seven security releases on July 23 after researchers published authenticated RCE PoCs for stock Redis 6.2.22, 7.4.9, 8.6.4, and 8.8.0. All four chains require RESTORE. The Streams chains also need EVAL and XGROUP; the 8.8.0 chain needs EVAL and the bundled RedisBloom module. Redis says the underlying memory flaws may lead to remote code execution . Redis 6.2.23, 7.2.15, and 7.4.10 fix the Streams shared-NACK use-after-free; Redis 8.2.8, 8.4.5, and 8.6.5 fix both the Streams issue and the RedisBloom and TDigest out-of-bounds writes; Redis 8.8.1 fixes the RedisBloom and TDigest loaders, while the Streams guard was already present in Redis 8.8.0. Two PoC targets, Redis 6.2.22 and 7.4.9, were the May security updates Redis told users to install, but those releases did not include the shared-NACK ownership guard. Upgrade to the fixed release for the deployed branch. Until then, revoke RESTORE from accounts that do not strictly need it and block untrusted network acce...

Cybersecurity researchers have uncovered a sandbox escape vulnerability in Anthropic's Claude Cowork that makes it possible to break out of the confines of a Linux virtual machine (VM) within which the agent runs to read or write files anywhere on the Mac. Accomplish AI, which shared details of the vulnerability with The Hacker News ahead of publication, said about 500,000 macOS users running local Cowork sessions were affected prior to it being patched. It has been codenamed SharedRoot . "We connected a folder to a fresh Claude Cowork session, sent one short message, and watched the agent escape the sandbox," Oren Yomtov, principal security researcher at Accomplish AI, said . "From inside the VM, it reached the host Mac and read and wrote files all over it, far outside the folder we'd connected, with no permission prompt anywhere." With this level of access, the agent can access any data stored on the Mac via the user's account, including SSH ke...

Most people understand identity theft as an attacker stealing a real person's sensitive information and impersonating them. Synthetic identity fraud is much harder to catch. Instead of stealing a real identity, the attacker manufactures a new one, frankensteining together several real data points with fabricated ones to create a person who doesn't exist. Since no real victim monitors misuse, a fake identity can silently accumulate permissions and credibility over time before it's ever detected. This same principle has a largely unexplored parallel with Non-Human Identities (NHIs). Security teams are spending significant effort to protect NHIs from being stolen. Still, the machine-side equivalent of synthetic identity fraud is rarely discussed: identities that were never legitimately provisioned from the start. Following this approach, an attacker doesn't hijack an existing service account but instead fabricates one, blending real environmental attributes with fake o...