Weekly AI, cybersecurity update: Microsoft, Google fix over 1000 bugs, Anthropic blocks bio-weapons supporting AI

Welcome readers! With this weekly series, we'll walk you through the top developments in AI and cybersecurity and explain why they should matter to you.

AI is changing the digital world rapidly. From enabling users to work efficiently to helping companies up their cybersecurity game to becoming a borderline threat, AI is actively help evolve the global cybersecurity landscape. The past week brought fresh evidence of this ongoing fundamental shift. On one hand we saw Google and Microsoft collectively patch over 900 vulnerabilities. On the other hand, AI companies reported incidents wherein their own models demonstrated unexpected hacking capabilities. At the same time, Anthropic disclosed an alarming incident wherein bad actors were trying to use its AI models for cyberattacks that could have supported biological weapons.

Here are the top developments you shouldn't miss. (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

For users this means that no amount of safety is enough. For businesses this means that protecting just the devices and the data isn't enough. Organisations must actively protect their AI models and put stringent guardrails to ensure their one of own agents doesn't go rogue.

So with this weekly series, we will walk you through the top developments in the world of AI and cybersecurity and explain why they should matter to you.

Google fixes 230 Chrome zero-day bugs one of which was exploited in the wild

Google recently confirmed that it has patched 230 vulnerabilities in its Chrome web browser. It has already started rolling out the security patch to all its Linux, Windows and Mac users globally. This security patch is available as a part of the Chrome 153.0.8010.36 for Linux, Chrome 153.0.8010.36 for Windows and Chrome 153.0.8010.37 for Mac. However, the catch is that Chrome users will get this update in the course of next days and weeks, which leaves room for these bugs to be exploited.

Google also confirmed that it was aware that one of the vulnerabilities mentioned in its list is a zero-day bug and that it had been exploited in the dark. This bug affects Chrome's V8 JavaScript and WebAssembly engine and it can allow a remote attacker to execute code inside Chrome's sandbox using a specially crafted webpage. The fix for this vulnerability is available in the Chrome 153 update.

"Google is aware that an exploit for CVE-2026-87491 exists in the wild," the company wrote in its Chrome security blog.

It is worth mentioning that this is the seventh actively exploited Chrome zero-day vulnerability that Google has patched this year. However, it hasn't disclosed any details about who could have exploited it and who could have been impacted by it.

Why readers should care: Chrome is one of the world's most widely used web browsers. So any vulnerability that can be exploited simply by getting a targeted user to simply visit a malicious webpage can have a large attack radius. To safeguard their data and their device, users should refrain from visiting any unknown or unsecure website and download the security patch as soon as it is available.

Microsoft fixes a record 974 vulnerabilities

Microsoft earlier this week confirmed that it had patched a whopping 974 vulnerabilities in its September 2026 Security update. Of these, 723 flaws have been fixed in Windows, 111 in Office and Office 2016, 62 in SQL, 12 in Azure and 22 in Developer Tools. The company also fixed two vulnerabilities, CVE-2026-85880 and CVE-2026-81963, both in Windows, that had been exploited by malicious hackers.

Why readers should care: Windows is one of the most widely used desktop-based operating systems in the world. So, any vulnerability could have a wide attack radius. Simply put, if you use Windows, Microsoft 365 or other Microsoft products, installing security updates regularly is one of the easiest ways to reduce your exposure to such cyberattacks.

Anthropic discloses fourth AI hacking incident involving Claude Opus 4.6

Anthropic, earlier this week, disclosed another incident in which one of its AI models breached external systems during testing. The incident involved an early version of Claude Opus 4.6, which dates back to January 2026. However, the issue wasn't discovered until August this year.

Anthropic, in a detailed blog post, said the model in question broke into third-party systems after failing to abort a task. The company also said that it had notified all affected organisations and subsequently expanded its investigation to roughly 481 million transcripts, involving external researchers to examine the behaviour of its models.

It is worth mentioning that this is the fourth such incident this year. Previous three incidents includes Mythos 5, Opus 4.7 and an unnamed model.

Why readers should care: This development is important as the wider concerns regarding AI are not limited to this technology being misused by malicious actors. More capable AI agents now can potentially execute tasks, interact with external systems and continue working even when something goes wrong. This not only ups the risks but also makes cybersecurity (externally) and internal guardrails more important than ever before.

Anthropic blocks attempts to misuse its AI to make biological weapons

Anthropic recently shared its Threat Intelligence Report 2026 wherein it disclosed several attempts to misuse its AI models for making dangerous biological weapons. The company highlighted five case studies, which included virologists working on a state-sponsored grant to pursue chikungunya gain-of-function work, a researcher in an unsupported region spending weeks planning avian influenza mammalian-adaptation experiments with Claude, a dozen customers using Opus 5 to draft a complete orthopoxvirus immune-evasion grant application, a state-supported researcher building a venom peptide atlas and generative optimization pipeline of molecules directed at paralytic and analgesic targets and a researcher computationally redesigning toxins for a national program, asking Claude to keep the agents’ identities deliberately vague in progress reports.

Anthropic said that it had blocked all of these requests and strengthened safeguards in newer models and blocked the accounts involved.

Why readers should care: This is one of the clearest examples of why AI safety is no longer just about misinformation or chatbot behaviour. While these models are beneficial for legitimate research, they can also be used for harmful purposes. So, putting more stringent guardrails has become more important than ever before.

OpenAI acknowledges it failed to disclose its earlier 'wiki incident'

OpenAI, as per a Reuters report, recently accepted that it did not initially disclose the earlier 'wiki incident'. For the unversed, the incident refers to a development wherein self-identifying OpenAI agents posted nearly 18,000 messages to a German public wiki where they discussed ways for other agents to bypass sandbox restrictions.

Responding to the development, OpenAI shared a statement on X wherein it said that its 'misalignment disclosure practices need to expand for this new phase of model capabilities'. "We and the larger AI community do not yet have a clear standard for how to report misalignment....We’re working on a framework and will share it in upcoming weeks, and in parallel we're working with dozens of government regulatory agencies worldwide on these issues," the company wrote in its post.

Why readers should care: As AI becomes more autonomous, the industry will need to set clearer standards around what constitutes an AI security incident and when companies should disclose it publicly. AI companies also need to strengthen their guardrails to avoid such incidents and more in the near future.

Google says attackers are using autonomous AI agents to automate entire attack chains

Google Threat Intelligence Group (GTIG) recently revealed that cybercriminals are moving from using AI as a coding assistant to using it to automate the entire threat chain. GTIG, in a recent blogpost, wrote that threat actors are now experimenting with autonomous, agent-enabled systems capable of planning and executing multiple stages of an attack. Such attacks reduce 'human-in-the-loop latency' and compress 'the traditional window for defenders to respond'.

Google said it observed a campaign wherein threat actors attackers had compromised a cloud resource and then used an AI agent to plan, build and execute a mass credential-harvesting operation in less than six hours. The company's researchers also identified attempts to manipulate AI coding assistants and security scanners as part of software supply-chain attacks.

Why readers should care: Automation is an important AI use case. However, when used in context of cybercrime, it can lead to more sophisticated attacks. It can also increase the scope of damage and reduce the window for companies to respond to such attacks. For end users, both individuals and companies, it means that the technology that they are using to automate IT and software development could also be used against them.