Microsoft Unveils AI Cybersecurity Solutions to Slash Costs

WHO:

CNBC - Microsoft CEO Satya Nadella spoke at the Microsoft AI Tour at TikTok Entertainment Centre in Sydney on April 23, 2026.

CNBC - Hayete Gallot rejoined Microsoft in February 2026 as executive vice president of security.

VentureBeat - Microsoft AI CEO Mustafa Suleyman discussed the new cybersecurity model and platform.

New York Times - Microsoft introduced MAI-Cyber-1-Flash and Project Perception.

WHAT:

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model, when paired with OpenAI's GPT-5.4, outperformed Claude Mythos 5 on the CyberGym benchmark, achieving a score of 96%.

VentureBeat - The new configuration of MAI-Cyber-1-Flash and MDASH is said to deliver roughly 50% cost savings compared to the current MDASH setup, which utilizes a blend of GPT-5.4, 5.4 mini, and 5.3 Codex.

VentureBeat - Microsoft processes more than 100 trillion security signals daily, consistent with its 2025 Digital Defense Report, which cited 4.5 million new malware files blocked and 5 billion emails screened per day.

VentureBeat - Suleyman stated, "We have trillions and trillions of data points going back decades," emphasizing the extensive data Microsoft has for training its models.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model achieved a score of 95.95% on the CyberGym benchmark, which is a known-vulnerability reproduction test.

VentureBeat - The model's deployment includes tenant isolation, auditing, and sandboxed execution environments with no internet access.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model is designed to handle up to 90% of security tasks efficiently, while the remaining 10% are escalated to OpenAI's GPT-5.4.

VentureBeat - The rollout of Project Perception is set to enter public preview on August 3.

VentureBeat - Microsoft is implementing strict access controls for the new model, monitoring API usage and requiring users to demonstrate good intent and technical competence.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model, when paired with OpenAI's GPT-5.4, outperformed Claude Mythos 5 on the CyberGym benchmark, achieving a score of 96%.

VentureBeat - The MAI-Cyber-1-Flash model is a sparse mixture-of-experts transformer with 137 billion total parameters and five billion active parameters, designed to handle up to 90% of security tasks.

VentureBeat - Microsoft states that the new configuration of MAI-Cyber-1-Flash and MDASH delivers "world-class performance at 50% of the cost of leading models."

VentureBeat - The rollout of the new model will be staged, starting with tens of approved users, then scaling to hundreds and eventually thousands.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model is designed to handle up to 90% of security tasks efficiently, while the remaining 10% are escalated to OpenAI's GPT-5.4.

VentureBeat - Suleyman stated, "We have trillions and trillions of data points going back decades," emphasizing the extensive data Microsoft has for training its models.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model achieved a score of 95.95% on the CyberGym benchmark, which is a known-vulnerability reproduction test.

VentureBeat - The model's deployment includes tenant isolation, auditing, and sandboxed execution environments with no internet access.

VentureBeat - Microsoft has stated that the model was evaluated by its AI Red Team and subjected to automated and expert-led adversarial exercises.

VentureBeat - Suleyman mentioned that Microsoft has used way less than 1% of its unique data set for training the model.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model, when paired with OpenAI's GPT-5.4, outperformed models including Mythos, Gemini, and GPT on the CyberGym benchmark, achieving a score of 96%.

VentureBeat - The rollout of Project Perception is set to enter public preview on August 3.

VentureBeat - Suleyman stated, "We have trillions and trillions of data points going back decades," emphasizing the extensive data Microsoft has for training its models.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model is designed to handle up to 90% of security tasks efficiently, while the remaining 10% are escalated to OpenAI's GPT-5.4.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model, when paired with OpenAI's GPT-5.4, outperformed Claude Mythos 5 on the CyberGym benchmark, achieving a score of 96%.

VentureBeat - Microsoft states that the new configuration of MAI-Cyber-1-Flash and MDASH delivers "world-class performance at 50% of the cost of leading models."

VentureBeat - The model's deployment includes tenant isolation, auditing, and sandboxed execution environments with no internet access.

VentureBeat - Suleyman stated, "We really do have a pretty significant data and harness and expertise moat, and that is enabling us to train models which are faster, better, cheaper, and I think this is genuinely the tip of the iceberg."

VentureBeat - Suleyman mentioned that there is a "massive pushback to reduce cost everywhere" as companies realize they are paying a "phenomenal amount of money" for AI models.

VentureBeat - Suleyman expressed skepticism about the industry's assumption that everything will converge into one giant unified model, stating, "It remains to be seen whether one giant model that is fully multimodal is actually able to deliver additional transfer learning benefit because of the integration."

VentureBeat - Microsoft is implementing strict access controls for the new model, monitoring API usage and requiring users to demonstrate good intent and technical competence.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model, when paired with OpenAI's GPT-5.4, outperformed Anthropic's Mythos 5, Google's 3.5 Flash Cyber, and OpenAI's GPT-5.5 Cyber on the CyberGym benchmark.

VentureBeat - The MAI-Cyber-1-Flash model is designed to handle up to 90% of security tasks, with the remaining 10% escalated to OpenAI's GPT-5.4.

VentureBeat - Microsoft states that the new configuration of MAI-Cyber-1-Flash and MDASH delivers roughly 50% cost savings compared to the current MDASH setup, which utilizes a blend of GPT-5.4, 5.4 mini, and 5.3 Codex.

VentureBeat - The rollout of Project Perception is set to enter public preview on August 3.

VentureBeat - Suleyman stated, "We have trillions and trillions of data points going back decades," emphasizing the extensive data Microsoft has for training its models.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model achieved a score of 95.95% on the CyberGym benchmark, which is a known-vulnerability reproduction test.

VentureBeat - The model's deployment includes tenant isolation, auditing, and sandboxed execution environments with no internet access.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model is a sparse mixture-of-experts transformer with 137 billion total parameters and five billion active parameters, designed to handle up to 90% of security tasks.

VentureBeat - The MAI-Cyber-1-Flash model achieved a score of 95.95% on the CyberGym benchmark, which is a known-vulnerability reproduction test.

VentureBeat - Microsoft states that the new configuration of MAI-Cyber-1-Flash and MDASH delivers "world-class performance at 50% of the cost of leading models."

VentureBeat - The model is designed to run inside MDASH, Microsoft's multi-agent harness for vulnerability identification and remediation, and is not available as a standalone public model or general-purpose API.

VentureBeat - The rollout of the new model will be staged, starting with tens of approved users, then scaling to hundreds and eventually thousands.

VentureBeat - Microsoft has stated that the model was evaluated by its AI Red Team and subjected to automated and expert-led adversarial exercises.

VentureBeat - The model's deployment includes tenant isolation, auditing, and sandboxed execution environments with no internet access.

VentureBeat - Suleyman mentioned that Microsoft has used way less than 1% of its unique data set for training the model.

VentureBeat - Microsoft claims that the MAI-Cyber-1-Flash model is designed to handle up to 90% of security tasks efficiently, while the remaining 10% are escalated to OpenAI's GPT-5.4.

VentureBeat - Microsoft processes more than 100 trillion security signals daily, which is consistent with its 2025 Digital Defense Report, citing 4.5 million new malware files blocked and 5 billion emails screened per day.

VentureBeat - Suleyman stated, "We have trillions and trillions of data points going back decades," emphasizing the extensive data Microsoft has for training its models.

CNBC - Microsoft has not disclosed the scale of its cybersecurity business since 2023, when it reported annual revenue exceeding $20 billion.

CNBC - Project Perception will enter public preview on August 3.

VentureBeat - Microsoft claims that MAI-Cyber-1-Flash is a 137 billion total parameter model with 5 billion active parameters, designed to handle up to 90% of security tasks.

VentureBeat - The MAI-Cyber-1-Flash model achieved a score of 95.95% on the CyberGym benchmark.

VentureBeat - Suleyman stated, "We have trillions and trillions of data points going back decades," emphasizing the extensive data Microsoft has for training its models.

CNBC - Microsoft claims its MAI-Cyber-1-Flash model outperforms Anthropic's Mythos 5, Google's 3.5 Flash Cyber, and OpenAI's GPT-5.5 Cyber on the CyberGym benchmark.

CNBC - Microsoft has not disclosed the scale of its cybersecurity business since 2023, when it reported annual revenue exceeding $20 billion.

CNBC - Microsoft introduced the Security Copilot assistant for cybersecurity practitioners in 2023, which incorporates OpenAI's GPT-4.

VentureBeat - Microsoft claims that MAI-Cyber-1-Flash outperformed Claude Mythos 5 on CyberGym, achieving a score of 96%.

VentureBeat - The new configuration of MAI-Cyber-1-Flash and MDASH is said to deliver roughly 50% cost savings compared to the current MDASH setup.

VentureBeat - Suleyman stated, "The whole game here is to reduce the costs. Mythos and so on are extremely expensive models... we want to be able to deliver better performance for cheaper."

VentureBeat - Microsoft claims that the new configuration of MAI-Cyber-1-Flash and MDASH delivers "world-class performance at 50% of the cost of leading models," emphasizing a shift towards cost-effective AI solutions.

VentureBeat - Suleyman noted that "the key barrier to adoption is access to chips, and cost is a function of chips," highlighting the limited supply of chips as a significant factor in AI deployment costs.

VentureBeat - Suleyman mentioned that there is a "massive pushback to reduce cost everywhere" as companies realize they are paying a "phenomenal amount of money" for AI models.

VentureBeat - Microsoft claims that MAI-Cyber-1-Flash and MDASH deliver "world-class performance at 50% of the cost of leading models," emphasizing a shift towards cost-effective AI solutions.

VentureBeat - The new configuration is said to deliver roughly 50% cost savings compared to the current MDASH setup, which utilizes a blend of GPT-5.4, 5.4 mini, and 5.3 codex.

VentureBeat - Microsoft launched MAI-Cyber-1-Flash, a custom-built cybersecurity model, and Project Perception, an agentic defense platform.

VentureBeat - MAI-Cyber-1-Flash scored 96% on CyberGym, outperforming models like Mythos, Gemini, and GPT.

Techmeme - Microsoft claims MAI-Cyber-1-Flash and MDASH deliver "world-class performance at 50% of the cost of leading models."

VentureBeat - Project Perception will enter public preview on August 3.

WHERE:

VentureBeat - The announcement was made by Microsoft, a global technology company.

WHEN:

CNBC - Project Perception will become available in public preview starting August 3.

VentureBeat - Project Perception is set to enter public preview on August 3.

WHY:

CNBC - Gallot stated that cybersecurity executives view the new model as a way to lower barriers and attract more talent to security operating centers (SOCs).

VentureBeat - Microsoft aims to reshape how enterprises purchase AI by focusing on cost-effective models.

VentureBeat - Suleyman stated that the goal is to reduce costs for customers, as companies are pushing back against expensive models.

HOW:

VentureBeat - The MAI-Cyber-1-Flash model runs inside MDASH, Microsoft's multi-agent harness for vulnerability identification and remediation.

CNBC - Microsoft is implementing a staged rollout for the new model, starting with tens of approved users, then scaling to hundreds and eventually thousands.

VentureBeat - Microsoft is implementing a staged rollout for the new model, starting with tens of approved users, then scaling to hundreds and eventually thousands, ensuring careful monitoring of API usage.

VentureBeat - MAI-Cyber-1-Flash is designed to handle up to 90% of security tasks, with MDASH escalating the remaining 10% to OpenAI's GPT-5.4.

VentureBeat - Microsoft processes over 100 trillion security signals daily, leveraging this data for cybersecurity insights.

VentureBeat - Microsoft is implementing strict access controls for the new model, monitoring API usage and requiring users to demonstrate good intent and technical competence.

(UPDATED July 28, 2026 19:23 UTC)

Microsoft has announced that its MAI-Cyber-1-Flash model, when combined with OpenAI's GPT-5.4, has outperformed Claude Mythos 5 on the CyberGym benchmark, achieving an impressive score of 96%. The company also revealed that this new configuration offers about 50% cost savings compared to the previous MDASH setup, which used a combination of different models. Additionally, the rollout of Project Perception is scheduled to enter public preview on August 3, and Microsoft is implementing strict access controls for the new model to ensure responsible usage.

(UPDATED July 28, 2026 18:21 UTC)

Microsoft has announced that its MAI-Cyber-1-Flash model, when combined with OpenAI's GPT-5.4, achieved an impressive score of 96% on the CyberGym benchmark, outperforming Claude Mythos 5. This new model, which features a staggering 137 billion total parameters and is designed to efficiently manage up to 90% of security tasks, will be rolled out in stages, starting with a select group of approved users. Additionally, Microsoft emphasizes that this configuration delivers "world-class performance at 50% of the cost of leading models," showcasing their commitment to innovation in AI security solutions.

(UPDATED July 28, 2026 16:28 UTC)

Microsoft has announced that its MAI-Cyber-1-Flash model, when combined with OpenAI's GPT-5.4, achieved an impressive score of 96% on the CyberGym benchmark, outperforming other models like Mythos and Gemini. The new configuration is also expected to provide around 50% cost savings compared to the existing MDASH setup, which uses a combination of GPT-5.4, 5.4 mini, and 5.3 Codex. Additionally, Microsoft processes over 100 trillion security signals each day and will launch Project Perception for public preview on August 3.

(UPDATED July 28, 2026 15:43 UTC)

Microsoft has announced that its MAI-Cyber-1-Flash model, when combined with OpenAI's GPT-5.4, achieved an impressive score of 96% on the CyberGym benchmark, outperforming Claude Mythos 5. This new model, featuring 137 billion total parameters and designed to tackle up to 90% of security tasks, is said to deliver world-class performance at half the cost of leading models, while also incorporating strict access controls and monitoring for user intent. Suleyman emphasized the importance of reducing costs in AI development, noting that companies are realizing they have been spending a phenomenal amount on these technologies.

(UPDATED July 28, 2026 14:43 UTC)

Microsoft has announced that its MAI-Cyber-1-Flash model, when combined with OpenAI's GPT-5.4, has outperformed competitors like Anthropic's Mythos 5 and Google's 3.5 Flash Cyber on the CyberGym benchmark, achieving an impressive score of 95.95%. This model is designed to manage up to 90% of security tasks, with the remaining 10% escalated to GPT-5.4, and it offers around 50% cost savings compared to the previous MDASH setup. Additionally, Microsoft plans to roll out Project Perception for public preview on August 3, further enhancing its security capabilities, which already process over 100 trillion signals daily.

(UPDATED July 28, 2026 13:27 UTC)

Microsoft has unveiled its new MAI-Cyber-1-Flash model, a sophisticated sparse mixture-of-experts transformer boasting 137 billion total parameters and designed to tackle up to 90% of security tasks. Achieving an impressive score of 95.95% on the CyberGym benchmark, this model is integrated within Microsoft's MDASH system for vulnerability identification and remediation, and it will be rolled out gradually, starting with a select group of approved users. Microsoft emphasizes that this new configuration delivers world-class performance at half the cost of leading models, while also ensuring robust security measures like tenant isolation and sandboxed execution environments.

(UPDATED July 28, 2026 11:40 UTC)

Microsoft has introduced the MAI-Cyber-1-Flash model, which is designed to efficiently manage up to 90% of security tasks, while the remaining 10% will be escalated to OpenAI's GPT-5.4. This new configuration, along with MDASH, is expected to provide around 50% cost savings compared to the existing MDASH setup, which uses a combination of GPT-5.4, 5.4 mini, and 5.3 codex. Additionally, Project Perception is set to enter public preview on August 3, marking another step in Microsoft's ongoing efforts in cybersecurity.

(UPDATED July 28, 2026 09:49 UTC)

Microsoft has unveiled its new MAI-Cyber-1-Flash model, which boasts a total of 137 billion parameters, including 5 billion active ones, and is designed to tackle up to 90% of security tasks. This advanced model achieved an impressive score of 95.95% on the CyberGym benchmark, showcasing its effectiveness in cybersecurity. Suleyman highlighted the extensive training data available for this model, stating, "We have trillions and trillions of data points going back decades," and noted that it operates within MDASH, Microsoft's multi-agent system for identifying and addressing vulnerabilities.

(UPDATED July 28, 2026 01:24 UTC)

Microsoft has announced that its new MAI-Cyber-1-Flash model outperforms competitors like Anthropic's Mythos 5 and Google's 3.5 Flash Cyber on the CyberGym benchmark, a significant development in the cybersecurity field. The company will begin a public preview of Project Perception on August 3, and it is implementing a staged rollout for the new model, starting with a select group of approved users before expanding to a larger audience. Hayete Gallot, who rejoined Microsoft as executive vice president of security in February, emphasized that this new model aims to lower barriers and attract more talent to security operating centers.

(UPDATED July 27, 2026 23:18 UTC)

Microsoft has announced that its MAI-Cyber-1-Flash model has outperformed Claude Mythos 5 on CyberGym, achieving an impressive score of 96%. The new configuration is expected to deliver around 50% cost savings compared to the existing MDASH setup, with Suleyman emphasizing the goal of providing better performance at a lower cost. To ensure a smooth transition, Microsoft will implement a staged rollout for the new model, beginning with a select group of approved users and gradually expanding to hundreds and eventually thousands, while closely monitoring API usage.

(UPDATED July 27, 2026 22:21 UTC)

Microsoft has announced a new configuration of MAI-Cyber-1-Flash and MDASH, claiming it delivers world-class performance at half the cost of leading models, which marks a significant shift towards more cost-effective AI solutions. Suleyman emphasized the importance of their extensive data and expertise, stating that while they can train models that are faster, better, and cheaper, the key barrier to widespread adoption remains access to chips, which directly impacts costs. He also mentioned a substantial push from companies to reduce expenses in AI, while expressing skepticism about the industry's belief in a single unified model delivering additional benefits through integration, and noted that Microsoft will begin a staged rollout of the new model, starting with a select group of approved users.

(UPDATED July 27, 2026 21:20 UTC)

Microsoft has announced that its new AI models, MAI-Cyber-1-Flash and MDASH, promise "world-class performance at 50% of the cost of leading models," marking a significant shift towards more cost-effective AI solutions. The new configuration is expected to deliver about 50% savings compared to the existing MDASH setup, which currently combines various versions of GPT technology. To ensure a smooth transition, Microsoft will implement a staged rollout of the new model, beginning with tens of approved users and gradually expanding to hundreds and eventually thousands, all while closely monitoring API usage.

(UPDATED July 27, 2026 20:22 UTC)

Microsoft has unveiled MAI-Cyber-1-Flash, a custom-built cybersecurity model, along with Project Perception, an agentic defense platform, aiming to transform how enterprises approach AI by offering cost-effective solutions. MAI-Cyber-1-Flash achieved an impressive 96% score on CyberGym, surpassing other leading models, and is designed to manage up to 90% of security tasks, while the remaining 10% will be escalated to OpenAI's GPT-5.4. Project Perception is set to enter public preview on August 3, with Microsoft emphasizing strict access controls and monitoring to ensure responsible usage of the new technology.