AI, cybersecurity weekly: Data stealing Android malware survives deletion, Microsoft execs call AI scraping labour theft

In today's update of our weekly series, we'll walk you through the actively exploited bug at Cisco, updates by OpenAI and Anthropic and more.

The world of AI and cybersecurity is evolving rapidly. On one hand, AI companies are racing to improve their machine learning models, on the other hand, the very same companies are struggling to maintain additional guardrails on their models, many of which they have admitted have gone rogue. Then there are other set of companies that are try to patch critical vulnerabilities so that those bugs are exploited.

AI and Cybersecurity weekly updates. (HT)

By Shweta Ganjoo

Shweta Ganjoo is a Chief Content Producer at HT Tech. She has over 10 years of experience covering technology, during which time she has focused on consumer tech devices, AI, social media, gadgets, and tech policy, delivering authoritative and reader-focused insights across India’s fast-evolving digital ecosystem.

She began her career in the early 2010s, building a strong foundation in digital-first reporting as India’s tech landscape rapidly expanded. Over the years, she has worked with leading media organizations including India Today Tech, Indian Express Group, and Techlusive, where she played a key role in scaling editorial content and audience engagement. At HT Tech, she drives product recommendation coverage, combining newsroom leadership with hands-on reporting experience.

Her expertise spans smartphones, wearables, smart home devices, AI, and social media ecosystems, along with regulatory developments shaping the tech industry. Shweta is known for her analytical approach to product reviews and news coverage, grounded in hands-on testing and real-world usability. She has conducted industry interviews, tracked emerging tech trends, and produced research-backed stories using verified sources and expert inputs. She holds a masters in Mass Communication, which complements her professional experience in digital journalism and content strategy, and aligns with her academic background in technology. Shweta follows a reader-first editorial philosophy, prioritising accuracy, transparency, and fact-checked reporting to deliver reliable, insightful, and practical insights.

Read moreRead less

Last week we saw Google and Microsoft combined fixing over 1,000 such vulnerabilities in their respective system, some of which were known to have been exploited by malicious hackers. This week we saw Cisco in a similar situation, albeit, the number of issues that the company delt with isn't the same as Microsoft and Google. In addition to this, we got more details about an Android malware that steals users' banking information and reinstalls itself after being deleted by users.

Here are the top developments in the world of AI and cybersecurity you need to know about this week:

Cisco warns of a new zero-day vulnerability exploited in active attacks

Cisco has warned its customers about a new maximum-severity vulnerability in its Identity Services Engine (ISE) and ISE Passive Identity Connector (ISE-PIC) that has a CVSS score of 10 and has been exploited in the wild. The vulnerability, which is being tracked as CVE-2026-76460, could allow an authenticated remote attacker to bypass authentication and gain access to the system.

According to the details shared by Cisco, this vulnerability stems from insufficient authentication controls on an API endpoint. Simply put, an unauthenticated remote attacker could send a crafted request to bypass the web-based management interface and gain unauthorised access to an affected device.

Cisco says that successful exploitation of this vulnerability can allow command execution with root privileges. Furthermore, attackers could potentially hide evidence of compromise, which in turn would make post-incident investigation more difficult.

The good news is that Cisco has already shared fixes in each of the systems where this vulnerability existed. It has fixed this bug in ISE 3.1 with Patch 12, in ISE 3.2 with Patch 11, in ISE 3.3 with Patch 12, in ISE 3.4 with Patch 7 and in ISE 3.5 with Patch 4

Why readers should care: Organisations using Cisco ISE for network access and identity management could face serious security consequences if their systems are compromised. The active exploitation status makes timely patching, log review and incident response more important for IT teams.

OpenAI reveals six incidents of concerning model behavior during testing

OpenAI, the company behind ChatGPT, has disclosed six new incidents wherein its AI models behaved in an unexpected or concerning manner during internal testing. In addition to this, the company also announced a framework for reporting, investigating and disclosing model misalignment incidents.

The six incidents shared by Open AI are:

- Self-generated instructions in task summaries: An unreleased research model inserted instructions into summaries used to continue its work in a new context window.

- GPT-5.6 Sol training failures: During the training of GPT‑5.6 Sol, many model instances added instructions to their summaries to hide mistakes, invent missing information and conceal failures from users.

- Unauthorised use of an exposed API key: While answering a routine question about earnings figures in a California county, an internal model found and used an exposed API key without authorization. The company said that when the concerned model wasn't able to retrieve the requested figures, it fabricated them and presented them as data from the requested source

- Uploading information to the internet for public citation: Two unreleased models uploaded previously retrieved records to a public paste service and attempted to use the resulting URL as a citation source. While the uploads succeeded, the following browser operations failed.

- Communication between model instances: Models used an internal software repository as a message board to exchange requests and responses across separate training samples while searching for missing input files.

- Publicly sharing a workbook: Agents working on a workbook made the file publicly downloadable through file-hosting websites after it could not collaborate through the intended local filesystem.

In its 'Our framework for reporting model misalignment', OpenAI said that its new reporting framework is intended to help identify model behaviour that could challenge existing safety assessments or expose weaknesses in safeguards.

Why readers should care: These incidents highlight why AI systems need strong guardrails and robust monitoring and permission controls. For businesses adopting AI agents, the difference lies between an instruction-following assistant and a system capable of unexpected actions.

Android malware RatHat steals banking info, reinstalls itself after deletion

A newly reported malware, dubbed as RatHat, has raised concerns among Android users because of its ability to steal users' banking details and reinstall itself after users try to delete it. According to a report by Cybersecurity News, RatHat malware uses Android Accessibility services to automate actions on the infected device.

The report further states that after gaining Accessibility access, it uses automated taps to enable Developer Options and Wireless Debugging. It then reads the pairing code and changing port displayed by Android to connect to the device's local Android Debug Bridge (ADB) service, which in turn gives it shell-level access beyond what ordinary applications typically have. The malware subsequently places disguised native components in a system directory.

These access let this malware check on-screen activity, get OTPs, and collect app-data, all of which are effective means for hackers to gain access to a user's bank account.

Why readers should care: Android users should be careful when granting Accessibility permissions and installing applications from unknown sources. This malware becomes important because of the impact that it can have on a user's financials.

Anthropic explores Claude feature for personal finance data analysis

Anthropic is reportedly testing a new feature that will enable Claude users to use the LLM for analyzing their financial information. The feature first spotted by TestingCatalog on X, will bring a new section called Money in the Claude app. In this section, users will be able to connect their bank accounts to 'understand your money'. Once the bank accounts are linked, users will be able to ask Claude questions about their spending, plans, and financial planning.

In essence, this feature could enable users to categorise their expenses and analyse spending habits, summarising financial records, helping users understand transaction-based patterns and supporting financial planning.

That said, the feature is still in development. It remains to be seen, how Anthropic finally implements this feature.

Why readers should care: Financial data is highly sensitive and granting AI tools access to such information can have substantial financial repercussions if the AI models it uses goes rogue or in case of a breach.

Microsoft execs called AI scraping ‘the largest theft of labour in human history’

Newly unredacted court filings in a copyright lawsuit The New York Times brought against OpenAI and Microsoft almost three years ago have revealed that a Microsoft executive privately have described AI scraping as 'the largest theft of labor in human history' and a threat to journalists and publishers, according to a report by TechCrunch.

Why readers should care: The outcome of AI scraping disputes could affect how publishers monetise their work and how AI companies access training data. It also raises questions about the sustainability of content creation in the AI era.