AI Companies Grapple with Cybersecurity Concerns for Large Language Models

AI companies face growing concerns over cyber threats to large language models

Artificial intelligence companies, including OpenAI, Microsoft, and Meta, are increasingly concerned about protecting their large language models (LLMs) from cyber threats. The focus is on protection, recovery, data portability, and data intelligence and working with other security companies to provide comprehensive solutions.Protecting large language models (LLMs) from cyber threats is the next big worry for artificial intelligence (AI) companies like OpenAI, Microsoft and Meta. Anand Eswaran, chief executive at data security and resilience firm Veeam Software, said that companies are "fairly worried" about security threats presented by AI, "but they're not thinking about everything they need to be doing in the correct way". "OpenAI, Meta, Microsoft and all the other companies that are creating these massive LLMs for public consumption have to think about it... what it means to protect the LLM so that criminals are not polluting it, to create hallucinations and bad outcomes," Eswaran said. "Public, open-source models and public consumer AI models will have to go through a journey. Otherwise, you will see corrupt models, leading to bad outcomes. It could be as simple as discrimination, which happens when you use an AI model for, let's say, loan approvals," he said in an exclusive chat with ET on the sidelines of the VeeamON India Tour event last month. Veeam Software offers data resilience solutions by providing data backup, recovery, security and intelligence to enterprise customers. Threats amplified According to a survey by Veeam, 75% of global organisations experienced a ransomware incident in 2023, and 81% of those chose to pay the ransom. But one in three organisations still could not recover their data, even after paying. Particularly in APJ (Asia Pacific and Japan) region, 29% of the organisations attacked could not recover data even after paying ransom. He highlighted that Microsoft 365 has 450 million users globally, of which, 55 million are currently protected. This offers billion dollars growth opportunity for Veeam. "We have a 40% market share in protecting Microsoft 365 from phishing and email threats, which are the number one-way companies get compromised. Only 13-14% of global Microsoft 365 users are protected, but this is changing." Founded in 2006, Veeam was acquired by Insight Partners in 2020 at a valuation of $5 billion. In April 2024, Veeam acquired ransomware incident response firm Coveware, followed by Alcion, an AI-driven data management startup which was acquired in September 2024. Veeam closed calendar year 2023 at a revenue of $1.5 billion and claims to be number one protection partner to Microsoft 365 covering 20 million users. "We will exceed $2 billion in revenue next year and continue a multi-billion-dollar journey," Eswaran said. The Seattle-based company started its India operations in 2014 and counts Hero MotoCorp, Mahindra Group and Future Generali as some of its customers today. It has 700 partners including resellers; alliances with global system integrators, independent software vendors, cloud providers etc. Pitfalls ahead India is among the top three growth markets accelerating at 45%, Eswaran said. "India is even further ahead in terms of digital transformation vision. Digital transactions, e-commerce, smart cities etc. are creating a massive amount of data across the board. India is important for me personally, as we are investing in people, resources, functions, and R&D." He explained that no single company in the security landscape can claim to be an end-to-end protection player. "Even if you look at the difference between a CrowdStrike, a Palo Alto, and a Splunk, they all deal with different aspects of the security stack. We work with all of them. What makes us special is that we don't claim to be an end-to-end security player. Our core focus is protection, recovery, data portability, and data intelligence." He said the Veeam has outperformed global leaders like Dell, Veritas and IBM. "In the last four years, we've gone from number four to number one. We took over as the number one market share leader in December 2022." Also Read: India's GenAI bet: going beyond LLMs?

AI companies need to be worried about cyber threats, model pollution

Artificial intelligence companies, including OpenAI, Microsoft, and Meta, are increasingly concerned about protecting their large language models (LLMs) from cyber threats. The focus is on protection, recovery, data portability, and data intelligence and working with other security companies to provide comprehensive solutions.Protecting large language models (LLMs) from cyber threats is the next big worry for artificial intelligence (AI) companies like OpenAI, Microsoft and Meta. Anand Eswaran, chief executive at data security and resilience firm Veeam Software, said that companies are "fairly worried" about security threats presented by AI, "but they're not thinking about everything they need to be doing in the correct way". "OpenAI, Meta, Microsoft and all the other companies that are creating these massive LLMs for public consumption have to think about it... what it means to protect the LLM so that criminals are not polluting it, to create hallucinations and bad outcomes," Eswaran said. "Public, open-source models and public consumer AI models will have to go through a journey. Otherwise, you will see corrupt models, leading to bad outcomes. It could be as simple as discrimination, which happens when you use an AI model for, let's say, loan approvals," he said in an exclusive chat with ET on the sidelines of the VeeamON India Tour event last month. Veeam Software offers data resilience solutions by providing data backup, recovery, security and intelligence to enterprise customers. Threats amplified According to a survey by Veeam, 75% of global organisations experienced a ransomware incident in 2023, and 81% of those chose to pay the ransom. But one in three organisations still could not recover their data, even after paying. Particularly in APJ (Asia Pacific and Japan) region, 29% of the organisations attacked could not recover data even after paying ransom. He highlighted that Microsoft 365 has 450 million users globally, of which, 55 million are currently protected. This offers billion dollars growth opportunity for Veeam. "We have a 40% market share in protecting Microsoft 365 from phishing and email threats, which are the number one-way companies get compromised. Only 13-14% of global Microsoft 365 users are protected, but this is changing." Founded in 2006, Veeam was acquired by Insight Partners in 2020 at a valuation of $5 billion. In April 2024, Veeam acquired ransomware incident response firm Coveware, followed by Alcion, an AI-driven data management startup which was acquired in September 2024. Veeam closed calendar year 2023 at a revenue of $1.5 billion and claims to be number one protection partner to Microsoft 365 covering 20 million users. "We will exceed $2 billion in revenue next year and continue a multi-billion-dollar journey," Eswaran said. The Seattle-based company started its India operations in 2014 and counts Hero MotoCorp, Mahindra Group and Future Generali as some of its customers today. It has 700 partners including resellers; alliances with global system integrators, independent software vendors, cloud providers etc. Pitfalls ahead India is among the top three growth markets accelerating at 45%, Eswaran said. "India is even further ahead in terms of digital transformation vision. Digital transactions, e-commerce, smart cities etc. are creating a massive amount of data across the board. India is important for me personally, as we are investing in people, resources, functions, and R&D." He explained that no single company in the security landscape can claim to be an end-to-end protection player. "Even if you look at the difference between a CrowdStrike, a Palo Alto, and a Splunk, they all deal with different aspects of the security stack. We work with all of them. What makes us special is that we don't claim to be an end-to-end security player. Our core focus is protection, recovery, data portability, and data intelligence." He said the Veeam has outperformed global leaders like Dell, Veritas and IBM. "In the last four years, we've gone from number four to number one. We took over as the number one market share leader in December 2022." Also Read: India's GenAI bet: going beyond LLMs?