Kategorie

Cybersecurity researchers have called attention to a new modular malware called TELEPUZ that's been spreading via websites infected with ClickFix lures since late April 2026. "The malware is full-featured, lightweight, and modular," Elastic Security Labs researcher Cyril François said in a technical report. "While the number of C2 [command-and-control] domains is currently small, the daily Ravie Lakshmananhttp://www.blogger.com/profile/[email protected]

Cybersecurity researchers have called attention to a new modular malware called TELEPUZ that's been spreading via websites infected with ClickFix lures since late April 2026. "The malware is full-featured, lightweight, and modular," Elastic Security Labs researcher Cyril François said in a technical report. "While the number of C2 [command-and-control] domains is currently small, the daily

ClickLock Stealer, a new macOS infostealer, answers a victim's refusal by killing their apps on a loop until they hand over the login password. It arrives as a command pasted into Terminal, asks for the password behind a fake system dialog, and when the victim cancels, installs two LaunchAgents and quietly exits. At the next login, Finder, the Dock, Spotlight, Terminal, Activity Monitor, and Swati Khandelwalhttp://www.blogger.com/profile/[email protected]

ClickLock Stealer, a new macOS infostealer, answers a victim's refusal by killing their apps on a loop until they hand over the login password. It arrives as a command pasted into Terminal, asks for the password behind a fake system dialog, and when the victim cancels, installs two LaunchAgents and quietly exits. At the next login, Finder, the Dock, Spotlight, Terminal, Activity Monitor, and

Two leading members of the Scattered Spider cybercrime collective were sentenced to five years and six months in prison each for hacking Transport for London (TfL) in 2024. [...]

Google DeepMind CEO Demis Hassabis is pushing for the US AI industry to self-regulate, with the support of government, as a starting point for an international creating shared international standards. In a blog post, he called for a focus on artificial general intelligence (AGI) and national security. 

But it is precisely that focus on national security that may make the results of such an effort, assuming it happens, less than palatable outside of the US.

“The rapid progress we’re seeing in AI requires a new approach to testing frontier AI model capabilities that is dynamic, adaptable, and rigorous,” Hassabis wrote. “The US is well positioned, given its economic and technical standing, to take the first step in developing such a framework. It could establish a new Standards Body modelled on a federally overseen public-private partnership or self-regulatory organization, much like the Financial Industry Regulatory Authority (FINRA), with a board that includes independent leading technical experts and open-source representatives.”

He noted, however, that the funding would need to be substantial, and would most likely come from industry, to allow the new body to attract world-class technical talent and obtain the necessary compute resources for large-scale testing.

Hassabis proposed that the organization “be responsible for developing assessment protocols and working with appropriate federal agencies and the US National Labs to conduct testing in areas relevant to national security,” and that AI vendor participants be encouraged to adopt best practices such as publishing model cards with technical details, maintaining strong internal cybersecurity, vetting key personnel, and providing sufficient resourcing for safety and security research.

This is not the first time Hassabis has expressed worries about AGI. 

DeepMind was involved in an earlier US government initiative evaluating AI safety, alongside Microsoft and xAI (now SpaceXAI) working with the Center for AI Standards and Innovation (CAISI), a division of the US Department of Commerce. It allowed CAISI to conduct pre-deployment evaluations and targeted research to “better assess frontier AI capabilities and advance the state of AI security.”  

The rest of the world may have concerns

Analysts and consultants were mixed about the move, with most expressing concerns about whether an industry-focused group would prioritize the public’s best interests.

“Self-regulation is not viable because it implies everyone is able to regulate themselves and will do so in line with the best interests of the public. Most tech vendors don’t have the capacity to self-regulate. They would just prefer a set of rules within which they can operate,” said Gartner VP analyst Nader Henein. “For-profit organizations are required to do what is best for their shareholders, and external regulation ensures that those organizations are never in a conflict of interest where they have to choose between what is good for their shareholders and what is good for the public.”

And, said Sanchit Vir Gogia, chief analyst at Greyhound Research, given the international nature of AI models, an effort coordinated by the US government might alienate other countries. 

“National security is the proposal’s accelerator in Washington and its poison pill abroad: the framing that opens the only gate available at home invites foreign capitals to read the institution as an instrument of American strategy,” he pointed out.

“The map is already plural,” he said. “Brussels switches on enforcement powers over general-purpose models [starting in August 2026], London runs the AI Security Institute, and Beijing licenses on its own terms. California and New York have legislated for frontier models at home. The durable route is shared technical evidence with sovereign enforcement, sealed through mutual recognition rather than deference, with India and the other major non-Western markets holding authorship rather than seats.”

Gogia added that the rules enacted by even such a group may not address all of the key concerns of enterprise IT. A US government effort along the lines that Hassabis is proposing would result in testing that “sits close to intelligence and industrial policy, and those functions will not stay neatly separated. A model can pass every catastrophic-risk test and still fail the enterprise on privacy, reliability, and liability,” he noted.

Walmart’s former director of cybersecurity Steven Eric Fisher, who is now an independent cybersecurity consultant, said he found the proposal “well-intentioned, but it addresses a highly polarized topic at a time when commercial interests carry unprecedented political influence, which is not always applied benevolently.”

He added, “an exclusive US standard that is not globally respected or enforceable would likely fail to achieve its core purpose and would place US companies at a competitive disadvantage.”

Aman Mahapatra, chief strategy officer for Tribeca Softtech, a New York City-based technology consulting firm, said that a deep dive into how FINRA operates today is illustrative of what IT leaders can expect from this effort, assuming the industry adopts that model.

“When the CEOs of the five companies that would be regulated are also the primary drafters of the standards, the standards will reflect those companies’ interests. FINRA has an independent board, but the operational reality is that member firm perspectives dominate the working groups that write the actual rules,” he said. “There is no reason to expect an AI equivalent to work differently, and every reason to expect it to work worse, because AI standardization is happening faster than any industry has ever attempted to standardize itself, and speed is the enemy of independent oversight.”

Carmi Levy, an independent technology analyst, was even more emphatically opposed to the Hassabis proposal.

“Asking Big Tech companies to self-police is analogous to allowing foxes to guard the henhouse. It hasn’t worked to date, and it won’t work going forward. Expecting these organizations to somehow change their ways at this point in time represents the height of naïve thinking,” Levy said. “The framework proposed by Demis Hassabis is a self-serving roadmap for an industry bent on racing to the AI horizon regardless of the harms caused along the way. It is impossible to quantify the dangers to broader society should frameworks allowing self-regulation become the norm.”

Some love the proposal

An almost completely opposite stance came from Yuri Goryunov, CIO of consulting firm Acceligence, who applauded the proposed move.

“This is one of the rare setups where industry self-regulation has a real shot, and enterprise IT should be enthusiastically rooting for it,” he said. “It fails when harms are externalized, such as in social media content moderation. Or when the overseer outsources judgment to the overseen, such as the FAA’s delegation to Boeing before the 737 MAX. It works when everyone in the industry shares the catastrophic downside.”

He suggested, however, that the best precedent here isn’t FINRA, it’s INPO, the Institute of Nuclear Power Operations, which the nuclear industry created within months of the 1979 Three Mile Island partial reactor meltdown “on the logic that an accident anywhere is an accident everywhere. INPO peer-reviews every US plant, its evaluations move insurance premiums, and it sits on top of the NRC’s statutory floor. That is a public-private stack very close to what Hassabis is describing. Frontier AI has the same structure: one lab’s catastrophic failure brings regulation down on all of them.”

For enterprise CIOs and other IT executives, Goryunov said, that model has the potential for being a big win.

“Today, every enterprise duplicates the same AI diligence of red-teaming, eval suites, governance committees and each does so with less information than any certifying body would have,” Goryunov said. “A credible standards regime does for AI what UL did for electrical equipment and SOC2 did for cloud: it converts an unknowable risk into a procurable product and gives boards a defensible standard of care. That’s not red tape. That’s peace of mind with an audit trail.”

However, Mahapatra said, “the countervailing view is that the alternative to industry-led standards is probably not thoughtful legislation. It is probably no standards, or state-by-state fragmentation, or the current pattern of ex-post enforcement actions where regulators surface concerns years after harm has already occurred.”

Thus, he noted, “Hassabis is making the reasonable argument that imperfect fast standards are better than perfect slow ones, and there is genuine merit to that view for topics like agent identity, evaluation methodology, and interoperability, which are exactly the areas OpenClaw is also targeting.”

This article first appeared on CIO.

The US Cybersecurity and Infrastructure Security Agency (CISA) has urged organizations to immediately secure Microsoft SharePoint deployments after warning that three vulnerabilities affecting the on-premises collaboration platform are being actively exploited.

A recent advisory from the federal cybersecurity watchdog asked administrators to patch vulnerable servers, review Microsoft’s mitigation guidance, and assume that internet-facing SharePoint instances remain attractive targets for attackers seeking an initial foothold into enterprise environments.

While applying patches remains the immediate priority, security experts caution that organizations should view the advisory as more than another Patch Tuesday exercise.

“This is what separates an IT incident from a business crisis,” said Chris Boehm, field CTO at Zero Networks. “One compromised SharePoint box is a ticket. That same box, with a clear path to your domain controllers, backups, and file shares, is how you end up with an encrypted infrastructure and a disclosure event. Segmentation stops the first from becoming the second.”

CISA’s advisory highlights CVE-2026-332201, CVE-2026-45659, and the newly added CVE-2026-56164, all of which have now been confirmed as exploited in the wild and added to the agency’s Known Exploited Vulnerabilities (KEV) catalog.

Exploitation tells a different severity story

The latest addition to CISA’s KEV catalog is CVE-2026-56164, an elevation-of-privilege vulnerability affecting Microsoft SharePoint Server. Although assigned a CVSS score of 5.3, the flaw can be exploited remotely without authentication, making it significantly more dangerous in practice than its severity rating alone suggests.

Microsoft has released security updates for supported SharePoint versions and recommended enabling the Antimalware Scan Interface (AMSI) integration to help detect malicious requests associated with exploitation attempts.

CISA also advised organizations to follow Microsoft’s incident response guidance, hunt for indicators of compromise, and rotate SharePoint machine keys where appropriate, acknowledging that patching alone may not fully remove attacker persistence from already compromised servers.

Older vulnerabilities remain active entry points

Alongside the newly disclosed flaw, CISA reiterated the urgency of addressing CVE-2026-45659, an insecure deserialization vulnerability allowing RCE that Microsoft had marked as “exploitation less likely” in its advisory in May. Another old bug CISA flagged is CVE-2026-32201, an improper input validation flaw that allows spoofing over a network.

Both of these flaws are being actively exploited in the wild.

CISA called out organizations failing to catch up with SharePoint updates, adding that attackers are increasingly targeting N-days rather than relying exclusively on newly discovered zero-days.

On concerns of patching speed, Boehm noted resilience is becoming an architectural challenge as much as an operational one.

“Stop measuring this in patch speed,” he said. “That’s a race you eventually lose. Some of these landed as zero-days with no fix on day one, and the window between disclosure and exploitation keeps shrinking. So the board-level question isn’t whether a server gets compromised. Assume one will. It’s how much of the business a single-owned system can take down with it.”

Boehm argued that limiting network reachability through segmentation should sit alongside patch management and threat hunting as a core defensive strategy. Reachability, he said, is a control that organizations own, not patch timing. CISA has given Federal Civilian Executive Branch (FCEB) agencies three days to remediate CVE-2026-56164 under Binding Operational Directive (BOD) 22-01.

The article originally appeared on CSO.

Introduction

In February 2026, we discovered a set of malicious activities that had been ongoing since late 2025. These activities involved a RAT module written in Go with proxy capabilities, which served as the main stage of the attack. The attack targeted government and diplomatic entities in Southeast Asia and showed a level of sophistication that caught our attention.

During the attack, the main malware, dubbed GoSerpent, received an encrypted argument and started communicating with a remote server. It was also used to deploy further malicious tools to collect sensitive data and dump credentials on the system.

Monitoring the activities of this threat actor revealed that in May 2026, they came back with an evolved set of malicious tools: a new RAT and proxy tool, Stowaway, which resembled the initial malware, as well as an additional stealthy tool to exfiltrate sensitive data collected in the previous few months through network shares.

We found earlier versions of the GoSerpent backdoor used since 2021 against victims in Southeast Asia with relatively simpler code that received command-line arguments in plain text. Even though the newer variant is stealthier, the attackers continued using the simpler version alongside the latest one in their recent attacks.

What makes this threat particularly concerning is the strategic deployment of various tools with sophisticated data collection and exfiltration capabilities.

In this article, we introduce the malicious tools uncovered by us, which have been used since late 2025.

Technical details Initial phase of the attacks

The initial phase of the attacks involved deployment of the GoSerpent backdoor, followed by additional malicious tools. During this phase, the main goal was to collect sensitive files and store them for future exfiltration, which was done by a data collecting tool, ThumbcacheService. The attackers also needed system credentials to exfiltrate the collected data through network drives at a later stage. This was achieved through a number of credential dumping tools deployed in this phase via the GoSerpent backdoor.

GoSerpent backdoor

The primary weapon in this campaign is the GoSerpent backdoor, a sophisticated Go-based remote access Trojan that has been active since at least 2021, with the most recent variant deployed in 2026.

This malware receives encrypted and base64-encoded command-line arguments containing a C2 server address and communication password, which are decrypted using AES-CBC mode with a fixed IV (31323334353637383930616263646566) and keys derived from predefined strings.

The backdoor connects to command-and-control servers using ChaCha20 encryption for communications, with the SHA256 hash of the communication password serving as the encryption key.

GoSerpent supports multiple C2 commands by receiving special command values. The commands include the following:

Command Symbol (as derived from corresponding function names) Description 2BA1 Sync Respond to the server to show the infection is active 3BA2 Exit Exit process 4BA3 Ls Start listening on a port 5BA4 Connect Connect to a remote server 6BA5 Hello Create a shell on the infected machine 7BA6 Ul Upload a file or directory to the server 8BA7 Dl Download from the server 9BA8 Ss5 Start a SOCKS5 proxy on the infected machine ABA9 Cl Close a listening port CBAB RF Forward to a connected node

GoSerpent can establish SOCKS5 proxy servers to route traffic through compromised hosts, enabling attackers to access other networks while masking their true IP addresses. The backdoor is capable of deploying additional malicious tools, including ThumbcacheService for file collection, Mimikatz for credential dumping, and QuarksDumpLocalHash for local account password hash extraction. The malware exhibits strong persistence mechanisms and uses filenames that mimic legitimate system processes such as lass.exe and updates.exe to evade detection.

McMx RAT

McMx is a basic Go-based proxy and remote access tool that represents a simpler variant of the GoSerpent backdoor, apparently compiled from a different GitHub repository path.

Unlike the latest variant of GoSerpent, which uses encrypted command-line arguments, McMx receives input parameters from text files in plaintext format — in a way that resembles older versions of GoSerpent. The malware features similar function names with apparent typos present in both tools.

Before executing McMx, attackers manipulate batch files to generate configuration files containing C2 parameters. The patterns observed show the use of echo commands to create configuration files with parameters like remote host addresses, ports, and secret keys. The McMx malware is then deployed with this configuration.

The tool shares core functionalities with GoSerpent, including:

• SOCKS5 proxying
• port forwarding
• file transfer
• remote shell capabilities

Data collection and credential dumping tools

Following initial deployment of the GoSerpent backdoor, attackers typically wait several days before utilizing it to download and execute additional malware components for data collection and credential dumping.

ThumbcacheService

ThumbcacheService is a malicious DLL deployed as a Windows service that functions as a sophisticated file collection mechanism within the GoSerpent ecosystem. The malware employs XOR encryption with a single-byte key of 0x13 for string obfuscation. It decrypts embedded strings and creates a database file named thumbcache_605a.db in the C:\Users\Public\ directory to store collected sensitive files. It specifically targets documents with the following extensions: .doc, .docx, .pdf, .xls and .xlsx.

The targeted files are then archived using 7-Zip and protected with a predefined password @vx0a9n5W2M0c3D6.#, enforcing a 20MB size limit for archives.
The malicious service also monitors the $Recycle.Bin directory for deleted files with the extensions of interest, ensuring comprehensive data collection.

Credential dumping tools

The threat actor deploys the following tools via GoSerpent backdoor to dump credentials:

1. Mimikatz — dumps memory from the LSASS process to extract credential material, including cached credentials and Kerberos tickets.
2. QuarksDumpLocalHash — extracts local account password hashes from the SAM registry hive, allowing for offline password cracking attacks.

These tools work together to maximize information extraction from compromised systems. The stolen credentials were used in later stages of the attack to facilitate the exfiltration of sensitive files collected by ThumbcacheService.

Second stage of the attacks

After the initial phase of the malware deployments, the attackers allowed a few weeks for the ThumbcacheService to silently collect sensitive files without exfiltrating them. In the meantime, the credential dumping tools also continued to steal credentials. In May 2026, the threat actor came back with a set of new tools. The main malware of this round of activity was another Go-based RAT and proxy tool, Stowaway. It was used to deploy the two-stage data exfiltration tool TmcLoader/TmcPayload, which was the last piece of the data theft puzzle.

Stowaway

Stowaway is a proxy and remote access tool compiled from an open-source framework with customized functions to make the infection stealthier. This malware features both network admin and agent capabilities, enabling attackers to establish chained proxy paths across multiple hosts with the following functionalities:

• SOCKS5 proxying
• port forwarding
• reverse tunneling
• remote shell access
• file transfer
• SSH-based tunneling

Communications are transported over TCP, HTTP, or WebSocket channels protected by AES-256-GCM or TLS encryption.
As the next step, the attackers deliver two files to the victim machine via Stowaway:

• TmcLoader with an embedded payload
• {BBF061R2-BE25-4F6D-8B2D-1A6A39C3FSA2}.db — an encrypted configuration file

TmcLoader/TmcPayload

TmcLoader is a stealthy C++ loader module registered as a Windows service. The malware embeds an encrypted payload dubbed TmcPayload within its .data section, which is decrypted and loaded into the memory space of the svchost process to maintain persistence and avoid detection.

TmcLoader employs dynamic API resolution through a circular XOR encryption, where each byte is XORed with the value of the subsequent byte, combined with Base64 encoding for string obfuscation to hide API names.

The loader creates a unique event to prevent multiple infections on the same system. After that, it extracts and decrypts the embedded TmcPayload. This payload component is responsible for exfiltrating sensitive data from the victim’s machine.

TmcPayload generates a file path from an obfuscated string: C:\Users\Public\Libraries\{BBF061R2-BE25-4F6D-8B2D-1A6A39C3FSA2}.db.

It then checks for the existence of this configuration file. If the file doesn’t exist, it delays execution for a random period of time before rechecking. The configuration file contains encrypted network share credentials and destination paths for data exfiltration. It specifically references the thumbcache_605a.db file created by ThumbcacheService as the file to be exfiltrated, demonstrating the integrated nature of the attack chain.

Toolset integration

What distinguishes this threat actor’s approach is the deliberate integration between different components of their toolset. The chain from ThumbcacheService to TmcLoader/TmcPayload demonstrates sophisticated operational planning:

1. ThumbcacheService: deployed via GoSerpent, collects and archives sensitive files into the thumbcache_605a.db database file.
2. Credential dumping tools: deployed via GoSerpent to retrieve system credentials.
3. Configuration file: delivered via Stowaway, contains credentials and file paths for data exfiltration.
4. TmcLoader/TmcPayload: deployed via Stowaway, reads the configuration file for data exfiltration.
5. Data transfer: using network credentials and destination paths from the configuration file, TmcPayload transfers the exact same thumbcache_605a.db.

This integration shows that the threat actor has carefully orchestrated their tools to work together seamlessly, ensuring that data collected by one component is available for exfiltration by another component.

Infrastructure

The malware operators leverage legitimate hosting providers, including Alibaba Cloud and UCLOUD HK, for their command-and-control infrastructure. The use of legitimate hosting platforms demonstrates operational security awareness, making detection more challenging.
The technical similarities between GoSerpent and the newer Stowaway tools strongly suggest the threat actor’s deep familiarity with network proxy technologies. The consistent use of legitimate domain names as secret keys, with GoSerpent employing www.microsoft.com and www.spacex.com and Stowaway utilizing github.code, indicates a standardized operational methodology.

Attribution

While the exact attribution of the GoSerpent campaign remains uncertain, there are indications of a potential link to the TetrisPhantom threat actor. The similarities in victim targeting, technical capabilities, and operational methodologies suggest a possible connection. However, further investigation is necessary to confirm this association.

Conclusion

The GoSerpent campaign represents a sophisticated and evolving threat to government and diplomatic entities in Southeast Asia. The threat actor’s use of customized tools, such as the GoSerpent backdoor, Stowaway, and TmcLoader, demonstrates a high degree of technical expertise and operational planning. The integration of these tools to collect and exfiltrate sensitive data highlights the actor’s focus on long-term access and intelligence gathering. As the threat landscape continues to shift, it is essential for organizations to remain vigilant and implement robust security measures to detect and prevent such attacks. By understanding the tactics, techniques, and procedures (TTPs) employed by this threat actor, defenders can better prepare themselves to counter similar threats in the future.

Indicators of compromise File hashes

GoSerpent
EBFFD5A76AAA690BCDB922F82E0BACC5
DC506FF7BB72735444FB3703A6BEE6D8

McMx
D6E86BF8A90E9B632ADD5FA495F97FBC

ThumbcacheService
CB6C4C70A3B171FA3404B8E1A3382116
64E9D1950E42BC98486DFD9919463D1C

Stowaway
CBBB6D483737EA3566726E51752DFF40
7F223EE0716CE2AD56F55D3744419449
19F8BEFCB035F52BF70094E6B4F5779A
846EF7C1C7323849B2A778C5E4CDA162

TmcLoader
D08A059E8B815E3B891505BC8777FC28
93A1569D5D5AB2C4761FEDF84F83709E

C2 IP addresses

152.32.160[.]239
8.220.194[.]108
8.220.214[.]132
8.220.209[.]155
8.220.193[.]189
101.36.104[.]87
144.48.6[.]46
103.138.13[.]30
47.80.22[.]58
152.32.222[.]113
43.106.30[.]226

Microsoft announced on Wednesday that systems running Windows 10 Enterprise LTSB 2016 and Home and Pro editions of Windows 11 24H2 will stop receiving updates in three months. [...]

More than 20 Brazilian government websites were hijacked and turned into malware delivery channels in an active PhantomEnigma campaign uncovered by ANY.RUN, a leading provider of interactive malware analysis and threat intelligence solutions. The investigation revealed previously undocumented backdoor behavior, hidden infrastructure relationships, and multiple attack arms behind a campaign

More than 20 Brazilian government websites were hijacked and turned into malware delivery channels in an active PhantomEnigma campaign uncovered by ANY.RUN, a leading provider of interactive malware analysis and threat intelligence solutions. The investigation revealed previously undocumented backdoor behavior, hidden infrastructure relationships, and multiple attack arms behind a campaign [email protected]

Ask an AI agent to summarize the reviews on a product page, and a single planted review can make it click "Buy Now" instead. Ask a coding assistant to apply a maintainer's fix from a GitHub thread, and a fake comment can make it run a stranger's command on your computer. Neither trick hijacks the agent's task. Each one just corrupts the facts it trusts and lets it carry on with the job you

Ask an AI agent to summarize the reviews on a product page, and a single planted review can make it click "Buy Now" instead. Ask a coding assistant to apply a maintainer's fix from a GitHub thread, and a fake comment can make it run a stranger's command on your computer. Neither trick hijacks the agent's task. Each one just corrupts the facts it trusts and lets it carry on with the job you Swati Khandelwalhttp://www.blogger.com/profile/[email protected]

An advanced malware previously attributed to a China-linked threat actor has resurfaced after more than four years within a Taiwan manufacturing firm, along with a previously unreported backdoor dubbed Stupig. Daxin ("srt64.sys"), as the kernel-mode rootkit is referred to, was first documented by Broadcom-owned Symantec in March 2022, with evidence indicating its use in targeted attacks aimed

An advanced malware previously attributed to a China-linked threat actor has resurfaced after more than four years within a Taiwan manufacturing firm, along with a previously unreported backdoor dubbed Stupig. Daxin ("srt64.sys"), as the kernel-mode rootkit is referred to, was first documented by Broadcom-owned Symantec in March 2022, with evidence indicating its use in targeted attacks aimed Ravie Lakshmananhttp://www.blogger.com/profile/[email protected]

Thinking Machines Lab, the San Francisco startup founded by former OpenAI CTO Mira Murati, has released Inkling, its first general-purpose AI model. The launch adds another US-developed entrant to an open-weight market where Chinese developers produce several leading coding and reasoning models.

Inkling uses a mixture-of-experts architecture with 975 billion total parameters, of which 41 billion are active during processing. It supports a context window of up to 1 million tokens and was pretrained on 45 trillion tokens spanning text, images, audio, and video. Thinking Machines said it also trained the model for coding, tool use, and multimodal tasks.

The release follows the October 2025 launch of Tinker, Thinking Machines’ first product and an API-based platform for customizing AI models. Developers can fine-tune Inkling through the platform.

In a June 2026 assessment, AI model routing platform OpenRouter highlighted DeepSeek V4 Flash, GLM 5.2, MiniMax M3, and Nvidia Nemotron 3 Ultra as four notable open-weight models. Nemotron was the only US-developed model in the group.

Performance and developer access

Thinking Machines Lab’s benchmark table shows mixed results. Inkling scored 77.6% on SWE-Bench Verified, behind DeepSeek V4 Pro and GLM 5.2 but ahead of Nvidia Nemotron 3 Ultra. It also recorded 74.1% on MCP Atlas, 77.1% on BrowseComp with context management, and 79.8% on IFBench.

Thinking Machines said Inkling’s result used a bash-only harness, while the comparison figures were reported by the competing models’ developers.

The model includes a reasoning-effort setting that developers can adjust from 0.2 to 0.99. Thinking Machines said the setting allows users to balance performance against the number of generated tokens. In the company’s testing, Inkling matched Nemotron 3 Ultra’s Terminal Bench 2.1 score while generating about one-third as many tokens.

Developers can fine-tune Inkling through Tinker using context lengths of 64,000 or 256,000 tokens and test it through the Inkling Playground. The model is available through APIs from Together AI, Fireworks, Modal, Databricks, and Baseten. It is also supported by inference software, including SGLang, vLLM, TokenSpeed, llama.cpp, and Hugging Face Transformers.

Inkling’s full weights are available on Hugging Face as the original checkpoint and as a quantized NVFP4 checkpoint. Thinking Machines also previewed Inkling-Small, which has 276 billion total parameters and 12 billion active parameters. The company said it would release the smaller model’s full weights after completing testing.

Enterprise impact

Inkling’s differentiation lies in its open weights, multimodal capabilities, controllable reasoning, and integration with Tinker, rather than benchmark leadership, according to Biswajeet Mahapatra, principal analyst at Forrester.

“Enterprises are most likely to benefit in workloads where domain adaptation matters more than generic model performance, including knowledge-intensive copilots, multimodal customer service, document understanding, operational workflow automation, and agentic tasks that require organization-specific data, policies, and processes,” Mahapatra said.  

Inkling’s US origin could also influence adoption among Western enterprises, according to Pareekh Jain, CEO of Pareekh Consulting. He said many Western organizations face regulatory or procurement barriers when considering Chinese-developed AI models.

“Inkling gives those organizations a US-developed open-weight option that they can deploy on their own infrastructure,” Jain said.

However, the benefits will need to be weighed against the cost of deploying the full model.

Running Inkling on private infrastructure requires a GPU cluster with at least 2 TB of aggregated VRAM for the BF16 checkpoint, according to the model card. Thinking Machines lists configurations of eight Nvidia B300 GPUs or 16 H200 GPUs. A quantized NVFP4 checkpoint lowers the requirement to at least 600 GB and can run on four B300 GPUs or eight H200 GPUs.

“Because Inkling is a massive model with 975 billion total parameters, running the full model still requires significant GPU infrastructure, making closed-model APIs more economical for many organizations,” Jain said.

Jain said Inkling-Small may be a more feasible option for many enterprises because it could reduce infrastructure costs and latency while retaining useful performance across key workloads.

Safety and governance

Thinking Machines said it trained Inkling for calibration, instruction following, and resistance to censorship. The company said the model showed “strong patterns of censorship non-compliance” when evaluated by Cognition on its Propaganda and Censorship Eval.

Inkling scored 98.6% on StrongREJECT, which Thinking Machines described as a test of whether models refuse unambiguous harmful requests.

The model’s safety behavior should be retested after an enterprise customizes it, according to Jain. “Model fine-tuning can weaken safety filters, so companies should retest safety after customizing the model rather than assuming it stays safe,” Jain said.

He added that self-hosted and modified versions could diverge from Thinking Machines’ official model over time without receiving automatic updates.

“CIOs need to ensure every AI agent action is logged, auditable, and governed by human approval for high-risk tasks,” Jain said.

The article originally appeared on InfoWorld.

CISA has ordered federal agencies to secure their systems by Saturday against ongoing attacks exploiting a critical vulnerability in the Oracle E-Business Suite financial application. [...]

A financially motivated Russian threat actor tracked as UAT-11795 is using trojanized software to steal credentials and cryptocurrency by deploying a new backdoor called Starland RAT. [...]

It’s not so much generosity that’s behind Anthropic’s decision to extend free access to its most advanced model, Fable, for paid subscribers until July 19, analysts say. Its a last-minute move to grab users, data and model evaluation results.

After the free-access period, Anthropic plans to convert Fable to a pay-per-use model, at $10 per million input tokens and a whopping $50 for 1 million output tokens.

That is double the price of its next most advanced model, Opus 4.8, for input and output tokens. “We’re extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19,” Anthropic’s team said in a July 12 tweet.

Anthropic keeps extending Fable because it does not yet know what its flagship is worth, said Sanchit Vir Gogia, principal analyst at Greyhound Research. “A vendor confident in its price does not move the same cutoff twice in six days, both times at the wire,” Gogia said.

Anthropic is essentially pushing deadlines to test its products, while users gain by being able to put their toughest tasks to Fable, Gogia said.

Anthropic, which did not immediately reply to a request for comment about the situation, has already seen plenty of action with Fable and its sister model Mythos. Both have been touted as the company’s most advanced models yet.

Fable stumbles, then reappears

Fable was officially launched June 9. Just three days later, on June 12, the US government put export controls on it after Amazon researchers bypassed Fable’s safeguards, prompting the model to identify software vulnerabilities and demonstrate an exploit. 

After Anthropic scrambled to address the issues — and after the export controls were lifted — Fable was relaunched July 1.

Fable’s freebie extension comes after OpenAI’s latest model, ChatGPT 5.6 Sol, became generally available July 9. Sol is cheaper at $5 per one million tokens input, and $30 for 1 million output tokens.

Anthropic and OpenAI are competing aggressively to build market share, said Jack Gold, principal analyst at J. Gold Associates. “Anthropic and OpenAI are looking to go public and the more users they have, the more attractive it is — even if they are not yet producing income,” he said.

In some ways, the two companies are following a well-trodden path to get customers hooked on their products and turned into paying customers. That’s what Meta, Google and Microsoft, for instance, have done over the years with various “free” offers that later morphed into paid products. 

Plus, said Gold, ”The more users you have, the better you can train your models across multiple data sets.”

That’s a potential boon for proprietary large language model (LLM) vendors offering free tokens in a bid to lock enterprises and vendors into their AI environments. But numerous experts have warned enterprises not to fall for that tactic. Instead, they argue enterprises should diversify AI development across multiple AI and cloud vendors, and adopt open-source models.

An LLM space race?

According to LLM benchmarks maintained by Artificial Analysis, Fable is the most intelligent model currently available, with Sol just behind it in second place. One benchmark by LiveBench places Sol as being better in reasoning, with Fable better at math, data analysis, instruction following and language. Both models have advantages in coding.

Meanwhile, Cursor and SpaceXAI on July 8 unveiled Grok 4.5, which the companies said can “handle difficult, long-running tasks that require creatively using tools to solve problems, whether in software engineering, data science, finance, legal work, or anything else you do on a computer,” the company said in a blog entry.

Its pricing is even more aggressive than Fable and ChatGPT 5.6 Sol. Grok 4.5 charges $2 for 1 million input tokens and $6 for 1 million output tokens.

There are growing concerns about tokenmaxxing, where enterprises rack up billions of dollars in token spending, blowing past usage limits before finance controls are implemented.

Enterprises might decide to spend more on models such as Mythos and Fable — if the benefits are tangible, said Max Leaming, head of data science and AI solutions at ManpowerGroup. Fable and Mythos may “actually be less expensive to use in spite of the spiked token cost because it’s far more efficient,” he said.

A company might find that the models use fewer tokens, are faster, and can reduce compute time, he said. “Even though the per-token costs may go up, we may see overall costs go down,” Leaming said.

Artificial intelligence (AI) is changing offensive security, but it has not changed the standard that matters most: a finding has to be proven before it becomes useful. AI-assisted tools can read code quickly, generate payloads, summarize attack surfaces, explain unfamiliar APIs, and run repetitive testing workflows at impressive speed. That is a real advantage for security teams. It also