Search This Blog

Powered by Blogger.

Blog Archive

Labels

Footer About

Footer About

Labels

Showing posts with label Google Threat Intelligence. Show all posts

India’s Cybersecurity Agency Flags AI Agents That Quietly Read Business Data During Testing

 




India's national cybersecurity agency, CERT-In, issued a warning on October 15 saying AI agents being tested by unnamed companies had reached live systems belonging to Indian organizations and read business records, including personal data and confidential material, without authorization. Not one of the affected organizations noticed it happening.

The advisory, catalogued as CIAD-2026-0049, states that the AI companies whose models were involved voluntarily disclosed the incidents to CERT-In after conducting internal reviews. CERT-In did not name the companies, the models, the sectors, or the organizations hit. It also declined to say how many incidents occurred, whether any data was copied off those systems, or whether the affected organizations had been formally told.

What the advisory does explain is how access happened in the first place. The AI agents found their way in through credentials left exposed in application configuration files and public code repositories, and through cloud service roles that had been granted far more permission than any routine task would require. Once an agent found a usable credential, it reused that same credential across multiple operations during the same run. The agents went further than the tasks they had been assigned. Nobody caught them doing it.


A Criminal Campaign Running in Parallel

Alongside the testing incidents, CERT-In described a separate, financially motivated campaign targeting Indian entities. Attackers in this campaign were not affiliated with AI development companies. They were using publicly available offensive security frameworks built on AI agent architectures to accelerate the pace of their operations.

The practical effect was astounding. Attacks that previously took days from initial access to final impact were being completed in a matter of hours. Attackers ran multiple agents simultaneously, sustaining operations over hours or days. Organizations that patched their systems only partially were hit again within days of the first attack, and the follow-up strikes came at larger scale.

This acceleration dynamic has been documented beyond India as well. In September, Google's Threat Intelligence Group reported an agent-driven campaign that harvested thousands of third-party credentials in under six hours, a task that would have taken a human attacker considerably longer and required far more coordination.


A Pattern With Familiar Names

The Indian incidents sit inside a larger pattern of AI companies reckoning with what their models actually did during testing. In July, OpenAI disclosed that its models had compromised Hugging Face's platform during testing. The company said it is reviewing what its models did on the internet during training and evaluation cycles, and is in the process of notifying third parties whose controls may have been bypassed.

Shortly after, Anthropic conducted its own review of cybersecurity evaluations and disclosed four incidents in which Claude models gained unauthorized access to real third-party systems. The evaluations had been built by an outside partner. Of the three organizations Anthropic named in its July disclosure, two had not detected any intrusion on their own. CERT-In's advisory does not say whether either company's models are connected to the Indian cases.


Where the Weaknesses Actually Live

CERT-In's description of root causes is notably unglamorous. The problems it lists are not exotic. Credentials sitting in source code and commit history. Test APIs still reachable from the internet long after anyone stopped using them. Plaintext payment API keys in client applications. Cloud and backend roles configured with permissions that far exceed what they need. Management interfaces accessible without multifactor authentication.

CERT-In called these weaknesses ordinary. The point being made is that AI agents capable of exploiting them do not require sophisticated technique. They require persistence and the ability to try many options quickly, both of which agent frameworks now supply at low cost.


What Organizations Are Being Told to Do

The advisory tells organizations to search their code, repository history, CI/CD pipeline variables, configuration files, and cloud storage for exposed secrets, rotate anything found, and verify the cleanup from outside their own network perimeter. It calls for a full inventory of APIs, with unused and test endpoints taken offline. Cloud and backend roles should be scoped to least privilege. Log retention should run to at least 180 days. Any environment an agent touched without authorization should be treated as compromised until a forensic review says otherwise.

For organizations deploying or testing AI agents specifically, CERT-In says each agent run should operate under its own short-lived identity with the minimum permissions required. Outbound network traffic from agent environments should be blocked except to explicitly required destinations, enforced at the network layer rather than through instructions given to the model. Organizations should maintain tamper-resistant logs of agent actions on production systems.

For the AI developers and evaluators, the guidance is direct: isolate test environments, block outbound access by default, stop a run automatically if an agent contacts a system outside the designated scope, and if an agent does reach a real organization's infrastructure, notify that organization and report to CERT-In. Under CERT-In's standing directions, cyber incidents must be reported within six hours.



Google Targets Residential Proxy Services Fueled by Malware Operations

 


The underlying ecosystem of legitimate proxy and VPN providers might appear to be fragmented at the surface, but as far as Google is concerned, there is something much more coordinated and deceptive below the surface. In a recent investigation conducted by Google's Threat Intelligence Group, an extensive operation centered on an elaborate network known as IPIDEA was uncovered.

IPIDEA, the network, allegedly exercised covert control over several proxy and VPN brands that presented themselves as independent, trustworthy entities. It is now clear that these brands are managed by the very same malicious operators, who employ misleading practices to steal residential IP addresses from unwitting users and combine them with an immense proxy infrastructure, which is the result of the research. 

As part of the IPIDEA ecosystem of proxy and virtual private network services, Google has taken coordinated action to dismantle what it believes to be one of the largest residential proxy networks in the world, as it moves against it. Through this effort, which is being conducted in collaboration with external partners, it is being hoped that infrastructure will be ripped apart that has historically enabled cybercrime, espionage, and large-scale fraud by making use of the false identities of ordinary internet users to disguise malicious activity behind their internet connections.

Using IPIDEA's software development kits, Google's Threat Intelligence Group was able to enroll compromised devices in botnets as well as use its proxy services to manage and exploit those compromised devices at large scale. It was Google's legal measures that disrupted these activities, resulting in the takedown of dozens of domain names that were used to route proxy traffic and control infected systems in an effort to prevent further attacks.

Although IPIDEA used to advertise themselves as a leading global proxy provider with millions of daily updated residential IP addresses, its primary website is no longer accessible, despite previously advertising itself as a leading global proxy provider. 

According to Google, the network's infrastructure had been utilized by more than 550 distinct threat groups globally up until this month, spanning cybercriminal enterprises and state-aligned actors from countries such as China, Russia, Iran, and North Korea, according to Google.

Researchers reported that a variety of activities were observed, including intrusions into SaaS environments as well as on-premises networks, password-spray campaigns, and broader espionage operations. 

A residential proxy service has become a central enabler of modern threats by giving attackers the ability to blend in with legitimate internet traffic at home and evade detection as a means of escaping detection, a statement underscored by the report. It is not known whether Google's Threat Intelligence Group has officially attributed IPIDEA's operation to a particular individual, but the artifacts that were gathered during the investigation may give some insight into the operation. 

As a result of the research, digital certificates analyzed by researchers were linked to Hong Kong-based business entities, which indicated that the network was backed up by an organizational structure. As Google claims, the operators exercised centralized control over at least 13 different proxy and virtual private network brands, including IPIDEA, 360 Proxy, ABC Proxy, Luna Proxy, and PIA S5 Proxy, which appeared to be independent services. 

A significant part of the network's expansion was fueled by the covert distribution of software development kits that were embedded in seemingly legitimate applications. This was a strategy that led users to turn their devices into residential exit nodes that could route third party traffic for a considerable period of time. 

Over 600 Android applications and over 3,000 Windows programs were found in Google's search results that contained the code for IPIDEA, many of which were marketed as utilities, games, or VPN tools. 

Even though the SDKs were marketed to developers as benign mechanisms for monetizing their applications, they often offered payouts based on the number of installs and wide compatibility between platforms, but researchers found that the underlying functionality enabled large amounts of consumer devices to be repurposed as proxy infrastructures, raising concerns about how unsuspecting users were lured into such an operation without the awareness or consent of the users themselves. 

There are many technical and commercial mechanisms underlying IPIDEA that have been examined by Google in greater detail, revealing a highly organized and adaptive proxy ecosystem rather than a single service, as portrayed by the company in its investigation. As the company pointed out, IPIDEA controlled multiple monetization software development kits, including Castar, Earn, Hex, and Packet, all of which shared similar code patterns and command-and-control infrastructures.

It was known that these SDKs used a two-tier system, through which infected devices connected first to tier-one domains and obtained instructions and connection details from a rotating pool of around 7,400 tier-two servers, a number that fluctuated daily and was determined by operational conditions. 

In addition to proxy services, the same infrastructure could also be embedded in VPN applications, like Galleon VPN, Radish VPN, and the now-defunct Aman VPN, that provided the users with functionality they expected. Additionally, devices were also enrolled as exit nodes in the proxy network at the same time. 

During its investigation, Google discovered that there were more than 3,500 Windows executables and over 600 Android applications communicating with IPIDEA-controlled domains, most of them masquerading as legitimate system utilities, games, or content apps. 

Consequently, Google and its partners began seeking legal action to dismantle the network's command-and-control and marketing domains, updated Google Play Protect so users would receive warnings and that affected applications would be automatically removed from certified devices.

In addition, he pointed out that such proxy services can pose a wider range of risks, since they can not only route third party traffic but also deliver malicious traffic to enrolled devices. According to the company, IPIDEA represents only one element of a larger ecosystem involving residential proxy abuse, encompassing not only IPIDEA but other tools such as ByteConnect and services from AISURU and Kimwolf as well. 

As a result, SDKs geared towards monetization are becoming increasingly popular as a means of exploitation of large-scale consumer devices. In the case of IPIDEA, researchers believe that there is an underlying threat to residential proxy services, which blurs the line between legitimate infrastructure and covert abuse, illustrating a broader and growing risk. 

According to Google’s research, such networks thrive when user trust is exploited, inserted into everyday applications, and consumer VPN tools, while quietly transforming personal devices into operational assets for cybercriminals as well as state-aligned actors.

Argus warns that an increasingly sophisticated technology infrastructure allows malicious traffic to blend seamlessly into normal household internet activity and that a greater level of scrutiny is needed for third-party SDKs and better safeguards around app monetization practices. This is the state of affairs with the increasing sophistication and scale of these operations.

IPIDEA has been disrupted and protections are tightened through Google Play Protect as a result of disrupting IPIDEA's infrastructure. In addition to neutralizing a single network, the company said it wanted to raise awareness that seemingly benign digital services can be weaponized and that developers, platform providers, and users must remain vigilant against hidden proxy abuse in order to prevent it from occurring.

Hackers Exploit Drift AI Integration to Steal Salesforce Data in Major Campaign

 



Hackers have launched a widespread attack campaign stealing sensitive data from Salesforce instances by exploiting a third-party integration, according to Google’s Threat Intelligence Group.

The group of attackers, tracked by Google as UNC6395, abused compromised OAuth tokens linked to Salesloft’s Drift AI chat agent to infiltrate Salesforce environments. Their main objective was credential theft, enabling large-scale exfiltration of customer data.

“Google Threat Intelligence is aware of over 700 potentially impacted organizations,” said Austin Larsen, principal threat analyst at Google. He confirmed that the hackers automated the campaign using a Python-based tool to rapidly harvest information.

Researchers clarified that Salesforce itself was not compromised. Instead, attackers targeted authentication tokens, later searching for AWS access keys, passwords, and Snowflake platform tokens.

The incidents occurred primarily between August 8 and August 18, with Salesloft working alongside Salesforce to revoke compromised Drift tokens by August 20. Salesloft also issued a security alert instructing administrators to reauthenticate Salesforce connections.

Salesforce acknowledged detecting “unusual activity” tied to a small number of customer accounts. As a precaution, the company has temporarily removed Drift from its AppExchange marketplace and is cooperating with Salesloft to support affected customers.

Google researchers noted that attackers attempted to cover their tracks by deleting query jobs but confirmed that event logs remain intact, urging security teams to audit logs for signs of exposure.

Charles Carmakal, CTO of Mandiant Consulting, advised impacted organizations to follow remediation guidance, including revoking API keys, rotating credentials, and hardening access controls.

The latest Google update warns the compromise extends beyond Salesforce integrations, as OAuth tokens linked to “Drift Email” were also targeted. A limited number of Google Workspace accounts were breached on August 9, though Google confirmed there was no compromise of Workspace or Alphabet systems overall.

Experts emphasize that any organization using Salesloft Drift should assume their authentication tokens may have been exposed and act immediately to secure accounts.