Search This Blog

Powered by Blogger.

Blog Archive

Labels

Footer About

Footer About

Labels

Showing posts with label Cloud Credentials. Show all posts

OVHcloud CLI Packages on GitHub Were Replaced With Secret-Stealing Malware

 



Attackers replaced the binary releases of OVHcloud's open-source CLI tool on GitHub with malicious builds designed to harvest secrets from developer machines, the company confirmed this week. The incident, which affected versions v0.11.0 through v0.15.0, went live on October 19 and was only caught after a customer raised an alarm about a suspicious release.

The malicious files were live for roughly 94 minutes. A customer's CI pipeline, which had a pinned SHA-256 checksum for the v0.15.0 Linux binary, failed when the hash no longer matched, and the customer opened issue #300 on the public GitHub repository. OVHcloud pulled the replaced assets at 14:21 UTC and shipped a clean replacement, v0.16.0, three hours later.


A Compromised Account, Months in the Making

The attack traced back to a single GitHub account: `HassenBF`, belonging to a legitimate OVHcloud contributor who had write access to the release repository. According to the security research group OpenSourceMalware, the account was compromised roughly seven months before the attack, with evidence pointing to early March 2026.

On March 5 and 6, four personal repositories on the account were force-pushed within a scripted window. Three of those rewrites embedded a `preinstall.js` loader file containing a payload hidden inside invisible Unicode variation-selector characters. A fourth repository inlined the same decoder into a React source file. The push cadence for the first three repositories was 42 seconds, which OpenSourceMalware interpreted as automated rather than manual activity.

After that, the account went quiet on the CLI project. It continued contributing to `ovh/ovhcloud-docs` through May to September 2026, with several pull requests merged normally. That cover of normalcy made the account hard to flag.


How the Malicious Binaries Were Built

The attacker did not touch OVHcloud's CI/CD pipeline. Instead, they built the infected packages locally and uploaded them directly using the GitHub Releases API, replacing the macOS (Intel and ARM), Linux (x86_64 and ARM64), and Windows x86-64 binaries across five versions. The i386, Windows ARM64, and WASM builds were left alone, which OpenSourceMalware read as deliberate targeting of developer and operations workstations.

Critically, the attacker also replaced the `checksums.txt` file last, so the new hashes matched the malicious binaries. Anyone who verified the checksum would still get a false sense of security.

OVHcloud CEO Octave Klaba confirmed that GitHub's release immutability setting had been turned off on the repository, which is what allowed an account with write access to silently overwrite already-published releases without triggering any CI checks. Since the incident, OVHcloud has re-enabled that setting. The v0.16.0 release is now marked immutable, meaning its files cannot be replaced.

A forensic check of the malicious binary using `go version -m` revealed it was built from a git checkout with uncommitted changes (`vcs.modified=true`), a dead giveaway that it came from a local workstation build rather than the official GoReleaser pipeline.


The Payload: A Solana Dead Drop and GlassWorm Fingerprints

OpenSourceMalware dug into the `preinstall.js` file loaded into the compromised account's repositories back in March. The loader uses a two-stage design. The first stage AES-256-CBC decrypts a hex blob, and the resulting second-stage script checks whether the victim machine appears to be based in Russia by testing language and timezone settings; if it is, the script exits after 10 seconds.

Otherwise, it contacts public Solana RPC endpoints, queries transactions from a specific wallet address, and reads a JSON memo embedded in the blockchain. That memo carries a base64-encoded URL pointing to the next-stage payload. This "dead drop" technique lets the attacker swap out the payload URL anytime without touching any code repository.

The Solana wallet in question, `BjVeAjPrSKFiingBn4vZvghsGj9KCE8AJVtbc9S8o8SC`, appears in public indicator lists for GlassWorm, a threat actor that has been running campaigns since late 2025 targeting VS Code, OpenVSX, GitHub, and npm. OpenSourceMalware attributes this attack to GlassWorm with moderate confidence.


Who Is Affected and What to Do

Any developer or CI pipeline that downloaded or ran the CLI through `install.sh`, the Homebrew cask, or a direct GitHub release between 12:47 and 14:21 UTC on October 19 may have been exposed. Klaba said the malware's goal was to collect secrets sitting on disk, and while he described the binary as low quality, he warned it was still capable of doing real damage.

Affected users should update to v0.16.0 or roll back to v0.10.0 immediately, but updating the tool alone is not enough. Any machine that ran a compromised version should be treated as having all its readable secrets exposed. That means rotating OVHcloud API keys, cloud credentials, SSH keys, Kubernetes configs, package registry tokens, and anything stored in `.env` files, from a separate, clean machine. Users on macOS should also check for an `~/init.json` file in the home directory, which the second-stage script writes as a throttle marker.

OVHcloud has said a full post-mortem is coming. Key questions remain open: how the developer account credentials were first stolen, whether the same actor maintained access for the entire seven months, and exactly what the final payload did before OVHcloud deleted the files.



AgentCorruption Exposes Security Risks in Amazon Bedrock AgentCore

 

A single prompt could have led to the exposure of conversations, source code, stored memories and cloud credentials across multiple AI agents in the same AWS account and region, according to Zenity Labs researchers. The so-called AgentCorruption attack chain targeted Amazon Bedrock AgentCore, which is AWS's managed service for building and running AI agents. 

The researchers' proof-of-concept involved an agent with a web-request tool that was prompted to call the local metadata endpoint at 169.254.169.254, which is where AWS workloads get temporary AWS credentials. Since the request came from inside the agent's Firecracker microVM, it returned credentials for the agent's execution role. 

This created a server-side request forgery (SSRF) vulnerability that was triggered by a prompt injection. "Exploitation didn't require a vulnerability in the chat interface; it required only that the agent have a tool to make the request and that the tool be permitted to call the metadata service," the researchers said. Zenity discovered that the broad permissions typically given to the default execution role allowed them to find agents, obtain container images, invoke other agents and access session events. 

They also discovered permissions that could be used to access conversations and implant false memories that would affect the agent's future responses. The researchers added that the latter capability could be used to implant "instructions" that would cause the agent to take specific actions. In addition, they found that these same permissions could be used to access API keys and secrets stored in AWS Secrets Manager, which could lead to lateral movement to other services and data. 

According to Zenity, metadata access was reported to AWS on Dec. 25, 2025, and excessive permissions in the execution role were reported on Jan. 12, 2026. AWS started the migration to Instance Metadata Service Version 2 in February 2026, and then made it mandatory. On Sept. 29, 2026, Zenity confirmed that AWS had removed the permissions that could be used to invoke other agents, chat with them or access the Secrets Manager. The researchers stated that the problems they reported had been addressed. 

AWS pushed back against the suggestion that this was a vulnerability, noting that it is normal for an agent to have access to its own execution role credentials via the metadata service. In addition, the company stated that cross-account access to agents is not possible without the appropriate permissions on the execution role and resources. 

Security teams should create their own IAM roles with limited permissions, block all outbound network connectivity except what is necessary, isolate agents that are accessible over the public internet and monitor logs in CloudTrail and CloudWatch for anything unusual, the researchers recommended. They also suggested that organizations ensure that the AgentCore Gateway cannot be bypassed to access agents directly. 

An AI agent's access to tools and permissions in the cloud can create security issues if not properly restricted, and this discovery shows why companies need to carefully control and monitor such access before actually launching an agent into production.

New Cuttlefish Malware Hijacks Router Connections, Cloud Data Stolen

 

In the ever-evolving landscape of cybersecurity threats, a new menace has emerged: Cuttlefish. This sophisticated malware targets enterprise-grade and small office/home office (SOHO) routers, posing a significant risk to both businesses and individual users alike. 

Discovered by Lumen Technologies' Black Lotus Labs, Cuttlefish operates by infecting routers and creating a proxy or VPN tunnel to stealthily exfiltrate data. By doing so, it bypasses security measures designed to detect unusual sign-ins, making it particularly insidious. One of the most concerning aspects of Cuttlefish is its ability to perform DNS and HTTP hijacking within private IP spaces. 

This interference with internal communications can disrupt organizational workflows and potentially introduce additional payloads, compounding the damage caused by the initial infection. While Cuttlefish shares some code similarities with HiatusRat, a malware previously associated with Chinese state interests, there is no definitive link between the two. Attribution remains challenging, further complicating efforts to combat this threat effectively. 

According to Black Lotus Labs, Cuttlefish has been active since at least July 2023, primarily targeting users in Turkey. However, infections have been reported elsewhere, impacting services such as satellite phones and data centres. The exact method of initial infection remains unclear, but it likely involves exploiting known vulnerabilities or brute-forcing credentials. Once inside a router, Cuttlefish deploys a bash script to collect host-based data and download its primary payload. 

What sets Cuttlefish apart is its adaptability to various router architectures, making it a versatile threat capable of targeting a wide range of devices. Once executed, the malware monitors all connections passing through the router, searching for specific data such as usernames, passwords, and tokens associated with cloud services like AWS and Digital Ocean. Once this data is captured, Cuttlefish exfiltrates it to the attacker's command and control (C2) server using a peer-to-peer VPN or proxy tunnel.

Additionally, the malware can redirect DNS and HTTP requests to actor-controlled infrastructure, enabling further data interception and manipulation. Cuttlefish severely threatens organizations worldwide, allowing attackers to bypass traditional security measures and dwell undetected within cloud environments. Network administrators should take proactive steps to strengthen their defences to mitigate the risk posed by Cuttlefish and similar threats. 

This includes eliminating weak credentials, monitoring for unusual logins, securing traffic with TLS/SSL encryption, and inspecting devices for signs of compromise. Additionally, regular router reboots, firmware updates, and password changes are recommended for SOHO router users to prevent exploitation.  

Cuttlefish represents a significant escalation in cyber threats, underscoring the importance of robust cybersecurity practices and constant vigilance in today's digital landscape. Organizations can better protect themselves against emerging threats like Cuttlefish by staying informed and implementing proactive security measures.

Sophisticated Cloud Credential Theft Campaign Targets AWS, Expands to Azure and Google Cloud

 

A cybercriminal group behind a sophisticated cloud-credential stealing and cryptomining campaign has recently expanded its targets beyond Amazon Web Services (AWS) to include Microsoft Azure and Google Cloud Platform (GCP). 

Researchers from SentinelOne and Permiso have been tracking the campaign and have found significant similarities between the tools used in this campaign and those associated with the notorious threat actor known as TeamTNT, who is primarily driven by financial motives.

The campaign's broader targeting started in June and has been evolving with incremental refinements since December. The recent attacks on Azure and GCP cloud services involve the same core attack scripts used in the AWS campaign. 

However, according to Alex Delamotte, a threat researcher at SentinelOne, the capabilities for Azure and GCP are less developed compared to those for AWS.

TeamTNT is well-known for exploiting cloud misconfigurations and vulnerabilities to target exposed cloud services. Originally focused on cryptomining campaigns, the group has now expanded its activities to include data theft and backdoor deployment. 

Recently, the attackers have been targeting exposed Docker services using modified shell scripts capable of profiling systems, searching for credential files, and exfiltrating them. They also collect environment variable details to identify valuable services for potential future attacks.

The attacker's toolset works across different cloud service providers and does not show significant automation for Azure or GCP beyond credential harvesting, indicating that much of the activity may involve manual intervention.

In addition to the shell scripts used in earlier attacks, TeamTNT has started using a UPX-packed, Golang-based ELF binary that drops and executes another shell script for propagating to other vulnerable targets. 

This worming propagation mechanism specifically targets Docker instances with certain user-agent versions, which could be hosted on Azure or GCP.

The researchers from SentinelOne and Permiso believe that TeamTNT is currently testing its tools in Azure and GCP environments without pursuing specific objectives on impacted systems. However, organizations using Azure and GCP should remain vigilant, as similar attack frameworks to those used against AWS may be employed against their cloud environments.

Recently, Sysdig also updated a report linking the ScarletEel cloud credential stealing and cryptomining campaign to TeamTNT's activity, further emphasizing the threat posed by this group. To defend against such attacks, administrators are encouraged to collaborate with their red teams to understand the most effective attack frameworks for these cloud platforms.

"Pacu is a known red team favorite for attacking AWS," she says. "We can expect these actors will adopt other successful exploitation frameworks."