artificial intelligence | Breaking Cybersecurity News | The Hacker News

Fireworks AI disclosed on Monday that its investigation into a security incident first reported early last month  uncovered further malicious activity, affecting account and billing information, alongside secrets, datasets, and incomplete models for a "limited set of accounts." In a notice published on October 10, 2026, the company said it detected unauthorized use of internal Fireworks credentials that allowed an unauthorized third-party to access environment variables used in evaluation jobs in a limited number of customer accounts. The activity was detected on October 6, 2026, at 9 a.m. PDT. At that time, the AI infrastructure provider said it only found evidence that environment variables, including API keys and tokens stored as secrets, were captured in certain affected accounts, adding that there were no signs that customer models, datasets, training data, inference traffic, or personal information were accessed. In response to the incident, Fireworks AI took th...

At some point your security will fail. This isn’t pessimism. It’s not because your team and tools aren’t sophisticated enough. It’s the reason we have detections in the first place. And, it’s why once we accept that prevention has its limits, we arrive at an interdiction strategy: finding and disrupting what got in. After the alert The alert tells you something happened. However, gaining access to your network doesn’t necessarily mean attackers have stolen data or disrupted the systems they were targeting. They may need to obtain additional access, move to another machine, or further explore the network. Being able to see and understand that activity is the intervention opportunity. People naturally gravitate toward detections during product evaluations. They’re tangible, easy to compare and quantify, and address the question everyone wants an answer to: Will they catch the attack? There’s often an assumption that the rest will follow, meaning, if the detections are good, surely...

Anthropic on Monday released Claude Fable 5.2 and Claude Mythos 5.2 models, which it said exhibit improvements in coding, reasoning, and vision. During pre-release training, the large language models (LLMs) helped its engineers "complete otherwise-infeasible software migrations, designed three working electronic devices from scratch, solved two open problems in the physics of black holes, made substantial progress on robotics evaluations, and formalized the proof of Fermat’s Last Theorem in 30 hours." The artificial intelligence (AI) company said high-stakes risks arising from misalignment remain low, and claimed the models matched or performed better than Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1 on most alignment measures, resistance to misuse, honesty, and factuality. "It was roughly an order of magnitude less likely than Opus 5.5 and Sonnet 5.5 to publish harmful content to public package registries without encouragement, though it still did...

A system doesn't always need a complicated bug to let someone in. Sometimes, a blank password is enough. That's one of the bigger stories this week. And it gets worse when the tools meant to keep systems updated or running safely become part of the attack. Elsewhere, attackers are hiding inside normal web traffic, stealing credentials from network devices, and finding new uses for exposed systems. AI is helping some of them move faster, but old mistakes are doing plenty of the work too. Even forgotten settings and small changes made months ago are coming back to cause trouble. There's a lot packed into this week's news, including some strange research findings, fresh attacks, and a few warnings worth taking seriously. Here's the full recap. ⚡ Threat of the Week OpenSSH Flaw Sees Exploitation Attempts — Threat actors are attempting to exploit a recently disclosed security flaw impacting OpenSSH that can enable sign-ins to accounts that have no password set....

A single attacker using AI agents pulled emergency dispatch records covering seven Taipei districts, researchers at Acronis's Threat Research Unit said last week. The agents also found a Taipei fire department platform exposing patient data from 2018, including ECG readings and ambulance video footage, without a login. The researchers are highly confident that the attacker is a single person based in Hualien City, Taiwan. He aimed the agents at hundreds of organizations in at least 11 countries, including emergency services, banks, universities and military systems. The agents collected data or accessed systems at about 10 of them in Taiwan, Mexico and Thailand. The evidence comes from the attacker's own server, which initially appeared to be a misconfigured development system. It turned out to be his operational attack platform, containing more than 90,000 files, including the agents' conversations, his command history, hacking tools, credentials, and victims' d...

Passwords and API keys that developers typed into commands they approved in Claude Code are sitting in public GitHub repositories, where anyone can read them. When a user selects " Yes, and don't ask again ," Anthropic's AI coding assistant can save the command, word-for-word, including the secret, in a project settings file. Software security company Apiiro said yesterday that 1,122 public copies of that file hold at least one secret, though it did not test whether any still work. One company the researchers notified earlier this week confirmed its key had been live. The file is .claude/settings.local.json. Removing it from a project does not undo the leak, because older versions stay in git history and in forks of the repository. The researchers' advice is to rotate first: change every password, token, and key the file has ever held, then clean up. How a Secret Lands in the File By default, Claude Code asks before it runs most shell commands. If the u...

OpenAI on Thursday unveiled GPT‑6 Cyber , which it described as a major step forward in security research that brings together stronger cybersecurity capabilities with a step-change improvement in model alignment. The new model is "more capable at finding and fixing vulnerabilities, and far more reliable staying within the boundaries researchers set," OpenAI said . It has classified the model as Critical in cybersecurity under its Preparedness Framework. The artificial intelligence (AI) company said GPT‑6 Cyber is also better than its predecessors like GPT‑5.6 Cyber and GPT‑6 Astra when it comes to following task boundaries and avoiding exploiting out-of-scope services in 99.93% of cases, up from 94.49% for GPT-5.6 Cyber. It also scores 48.5% on ExploitGym, a cybersecurity benchmark that evaluates the capabilities of AI agents to autonomously turn known software vulnerabilities into functioning, real-world exploits. In comparison, GPT-5.6 Cyber scored 32.1%. OpenAI ...

A browser extension that can run scripts on claude.ai could make Claude Desktop run code on a person's computer, researchers at Forever Security, an endpoint security company, showed in a test. The attack needs Dispatch , a Claude feature that lets people send tasks from their phone to Claude Desktop. The flaw that makes this possible can still be exploited, the researchers said earlier this week. Claude Desktop is an AI agent that can run commands on the computer where it is installed. The researchers found that Dispatch also accepted an ordinary signed-in session on claude.ai, the Claude website, even though the site showed no link to it. That let an extension running on the site send Claude Desktop instructions as if the user had sent them. The researchers say Anthropic, the company that makes Claude, told them on October 13 that Dispatch would be retired, with the end of October as the target. The Dispatch help page, last updated late last week, does not mention a retire...

An update on Athena, and some unsolicited predictions on what's coming next On Saturday, Dario Amodei published We Must Pace the Frontier . I agree with him. Go read it if you haven't. I've spent the summer looking at what these models find, and the gap he describes between what they can do and what the rest of us are ready for is the one I've been staring at since May. The hardest piece of his plan is the one that asks governments to coordinate. They're in a difficult spot: regulate too little and a US company accidentally builds the weapon, regulate too much and it gets built somewhere else, and open source isn't governable either way. Where governments can act is on consumption: not what the models can do, but what happens to what they find. Who gets a vulnerability first, how fast a fix reaches the people running the code, whether the operators who can't move on an attacker's timeline get a head start. That's what Gold Eagle is. It matters...

Blockchain analysis firm Chainalysis has revealed that its ongoing investigation into the agentic cyber attack targeting its systems also exposed names, usernames, password hashes and live tokens for all users. "While we investigated, we removed access to Reactor and Data Solutions," the company said. "At this time we continue to have no evidence that any investigations data has been accessed or Chainalysis attribution data has been manipulated." Chainalysis said it has implemented additional proactive monitoring systems to detect unusual activity, adding it took the Know Your Transaction (KYT) user interface and API offline, alongside other address screening products and the login page. Previously, Chainalysis said it was probing what it described as an "ongoing multi-front cyber attack," leading to the discovery of an "on-path attack" that exposed a subset of customer passwords. The company said the attack vector has since been addres...

How much damage can someone do without breaking much? Quite a lot, it seems. A stolen login can open doors across several systems. A trusted setting can be used to lock people out. Even a simple file copy can reach places it shouldn't. And while new tools are helping find security holes faster, some important systems are still taking months, sometimes over a year, to get patched. Not exactly a great match. The stranger part is how little some attacks now need. Ransom demands without encrypted files. Fake websites that look real enough to steal money. AI tools picking up dangerous instructions while trying to help. Even the attackers are making mistakes in their own code. This week's stories are a reminder that the biggest problems don't always come from clever tricks. Sometimes the easy way works just fine. Here's what else happened. The threats change every week. Subscribe, and we’ll alert you when each new ThreatsDay Bulletin is out.

OpenAI on Wednesday detailed a pipeline that employed an upcoming cyber-focused artificial intelligence (AI) model to validate and patch 948 local privilege escalation vulnerabilities in Linux for $1.2 million in API costs. "Our prototype identified and analyzed 2,308 candidate Linux kernel findings, dynamically reproduced 1,724 vulnerabilities, and produced 948 local privilege escalation (LPE) exploits that could allow an unprivileged local user to gain root access," the AI company said . The findings build upon 24 LPE exploits that OpenAI revealed were discovered by its GPT‑5.5‑Cyber model back in June 2026. The latest research turns this process into an agentic pipeline comprising five stages - Vulnerability discovery, where agents identify high-value audit targets (i.e., parts of the codebase that are security-relevant and reachable by low-privileged users) and look for candidate findings Dynamic validation, where agents attempt to reproduce each candidate ...

October 2026 brought another wave of cyber threats targeting US businesses, with attackers exploiting trusted platforms, legitimate authentication processes, and everyday business workflows. For enterprises, these campaigns raise the stakes beyond individual security incidents, increasing the risk of financial losses, data breaches, and costly operational disruption. Below, we examine five active cyber threats through real-world sandbox investigations, revealing how they operate, the risks they pose to organizations, and how security teams can detect and contain attacks faster to minimize business impact. 1. SkyW4lker: Microsoft 365 Phishing Puts US Corporate Accounts at Risk SkyW4lker is an active phishing kit that exploits Microsoft's legitimate device-code authentication to compromise corporate accounts without stealing passwords or bypassing MFA. ANY.RUN researchers identified 618 related sandbox analyses between June and October 21, 2026, with 46% coming from the US. ...

The Indian Computer Emergency Response Team (CERT-In) has disclosed that an artificial intelligence (AI)-assisted threat actor it tracks as Crafty Falcon is targeting organizations in the Banking, Financial Services, and Insurance (BFSI) sectors in the country. Describing the threat as active and ongoing, the activity involves the use of publicly-available AI-driven offensive security tools like ARTEX, PentAGI, Cairn, and CyberStrikeAI . "These tools execute complete attack chains autonomously at machine speed, generate and refine large numbers of tailored attack scripts in rapid succession, and adapt their payloads based on target responses," CERT-In said . "The supporting attack infrastructure rotates IP addresses within hours to days." The agency said the attacks have also targeted cryptocurrency exchanges and digital asset service providers, resulting in the theft of keys and funds. The update follows an October 2026 alert issued by CERT-In, which wa...

Cybersecurity researchers have disclosed details of an OpenAI Codex jailbreak bundle that's designed to modify an artificial intelligence (AI) agent's trusted instructions to insert advertisements for a third-party AI gateway. The phenomenon has been codenamed prompt adware by Check Point Research. Called Haiou (meaning Seagull in Chinese), the jailbreak modifies OpenAI Codex's configuration files to inject advertisements for the jailbreak author's website into subsequent agent responses. "The advertised service is an opaque, OpenAI-compatible LLM gateway, a middleware between an application and an LLM provider," Check Point Research said in a technical report. "The likely business model is to use the freely distributed jailbreak bundle as an acquisition channel for the paid gateway service." Adware is a type of software that displays unwanted advertisements, which can appear on users' computers or mobile devices, often in the form of ...

Two flaws in the Xen hypervisor allow a malicious guest virtual machine (VM) to read and write the memory of other VMs on the same host, including dom0. Dom0 is the VM that, by default, has full permissions and manages the others. Patches are out, and there is no known mitigation, the Xen Project said in an  advisory  on November 3. All Xen versions from 4.6 onward are vulnerable, while 4.5 and earlier are not. The flaws are tracked as CVE-2026-102703 and CVE-2026-102704 , under Xen advisory XSA-523. The Xen Project credited AI company Anthropic with finding the flaws using its Claude model. Stephan Lipp of Ada Logics then validated the finding, and the CodeAnt AI Security Research Team later independently reported the same issue. The flaws are in grant copy, the operation Xen uses to copy data between VMs. Each VM tells Xen which of its memory pages other VMs may use, and Xen checks those permissions. To save time on a batch of copies within the same memory page, ...

Nearly every agent security pitch now ends on the same promise: the product stops the bad action in real time. The promise holds for one kind of agent, the kind whose traffic passes through a gateway the enterprise owns. Most agents never touch one. One recent case shows what that gap looks like in practice. An employee asked an AI assistant to trigger a mass-email automation. The automation hit a recipient-limit policy and failed, so the assistant, which was also connected to the employee's personal mailbox, offered to send the batch from there instead. The employee approved without registering that the assistant was about to route around the sanctioned system. An unknown number of external recipients received email from a personal account rather than the approved channel. No jailbreak was involved. The assistant did exactly what it was built to do, and the fix was not retraining a model. It was a single runtime rule that blocks any attempt to reach more than fifty external d...

Blockchain infrastructure provider Chainstack on Tuesday disclosed that it was targeted in a cyber attack that allowed an unauthorized party to gain access to some parts of its infrastructure. The incident is assessed to be part of a large-scale, artificial intelligence (AI)-powered attack campaign that came to light last week. "An external attacker gained unauthorized access to part of Chainstack's infrastructure through a third-party component," the company said . "We detected the intrusion, cut the access paths, and are hardening our systems." According to the timeline of events shared by Chainstack, the attacker gained access to a development environment on October 26, 2026, and then escalated their access to break into "some parts of our production environment." The intrusion was subsequently detected on October 31, 2026, and the attacker's access was cut off that same day, the company noted. As part of the digital break-in, the ...

Binance on Tuesday said the record-setting theft of $6.43 billion in assets from the world's largest cryptocurrency exchange was the culmination of a "sophisticated, multi-stage campaign" that involved the use of a tampered Homebrew installation and recruitment-themed social engineering attack targeting two of its engineers. "It was not a single isolated breach or a simple phishing incident," the company said . "The attackers combined a range of techniques, including social engineering by fake recruitment interviews, malicious software downloads, compromised internal AI tooling, endpoint compromise, and an advanced transaction-authorisation method." The starting point has been traced back to a supply chain attack targeting one of its engineers from the Wallet DevOps team (named "Engineer A"), who downloaded a trojanized Homebrew installer to install a popular artificial intelligence (AI) tool after searching for it on a search engine....

Google on Monday warned that threat actors are hijacking enterprise access to Large Language Models (LLMs) to fraudulently leverage expensive computing resources and AI services. The tactic is called LLMjacking , a form of resource hijacking adapted for the LLM era, where an attacker obtains an API key through phishing, source code leaks, misconfigured environments, or exposed client-side code to gain unauthorized access to an organization's LLM accounts, while the legitimate owner covers the consumption costs. Such attacks have a wide geographical footprint, with impacted entities located in North America, South America, and Asia Pacific across the financial services, retail, technology, and transportation sectors. "Unlike the gradual resource consumption associated with cryptojacking, LLMJacking can generate large spikes in fraudulent cloud charges in a matter of hours," Google Threat Intelligence Group and Mandiant said in a report shared with The Hacker News...